1  

Dieter Fensel Holger Kett Marko Grobelnik (Eds.)

Common  Value  Management    

     

1st  International  Workshop  on  Common  Value  Management  CVM2012  

   

at  the  Extended  Semantic  Web  Conference  2012  (ESWC2012)  |  May  27-­31,  2012  |  Heraklion,  Greece  

  2  

Imprint  

 

Editors  Dieter  Fensel  (STI  Innsbruck,  Innsbruck,  Austria)  Holger  Kett  (Fraunhofer  IAO,  Stuttgart,  Germany)  Marko  Grobelnik  (Jozef  Stefan  Institute,  Ljubljana,  Slovenia)  

Authors  Dieter   Fensel,   Andreea-­‐Elena   Gagiu,   Harriet   Kasper,   Holger   Kett,   Iva   Koleva,  Birgit   Leiter,   Inna   Novalija,   Ioannis   Stavrakantonakis,   Tadej   Štajner,   Stefan  Thaler,  Andreas  Thalhammer,  Ioan  Toma  

Cover  Design  Carmen  Brenner,  Ioan  Toma  

Publishing  Fraunhofer  Verlag,  Stuttgart  

Sales  and  Distribution  Fraunhofer-­‐Institut  für  Arbeitswirtschaft  und  Organisation  IAO  Nobelstrasse  12,  70569  Stuttgart  Tel.:  +49  (0)711  970  5120  Fax  +49  (0)711  970  5111  presse@iao.fraunhofer.de  http://www.ebusiness.iao.fraunhofer.de  

 

 

 

Printing  and  Bindery:  

Mediendienstleistungen  des    Fraunhofer-­‐Informationszentrum  Raum  und  Bau  IRB,  Stuttgart    

Printed  on  acid-­‐free  and  chlorine-­‐free  bleached  paper.  

All  rights  reserved;  no  part  of  this  publication  may  be  translated,  reproduced,  stored  in  a  retrieval  system,  or  transmitted  in  any  form  or  by  any  means,  electronic,  mechanical,  photocopying,  recording  or  otherwise,  without  the  written  permission  of  the  publisher.  

Many  of  the  designations  used  by  manufacturers  and  sellers  to  distinguish  their  products  are  claimed  as  trademarks.  The  quotation  of  those  designations  in  whatever  way  does  not  imply  the  conclusion  that  the  use  of  those  designations  is  legal  without  the  consent  of  the  owner  of  the  trademark.    ©  by  Fraunhofer  Verlag,  2012,  ISBN  978-­‐3-­‐8396-­‐0411-­‐3  

  3  

Forword

In current times, next to high quality products or service, being competitive also means to be active on various channels of Social Media such as Blogs, Twitter, Facebook, Google+. This results in a huge workload which can be exemplified by considering small hoteliers. How can they ensure that they are found by their potential customer, i.e., how can they find them? They should have web sites with a high visibility on various search engines, they must be present in a large number of on-line booking channels, they should be found through the web site of the town, and obviously a Facebook site is a must. Also, bookings through mobile platforms are significantly increasing and many of them want to be found also in this domain. Why not add a video about the hotel on YouTube, a chat channel for instant communication, fast email and fax response capabilities, the old-fashioned telephone, and occasional tweets and emails that are clearly distinguishable from spam?

The 1st International Workshop on Common Value Management (CVM2012) focuses on one of today’s major trends, not only in relation to semantic technology research, but also in relation to the industries’ huge demand for tools and consulting in the domain of social media. The term Common Value Management (CVM) serves as an umbrella for three major aspects of today’s organization on-line communication:

• Yield Management also known as Revenue Management "is an economic discipline appropriate to many service industries in which market segment pricing is combined with statistical analysis to expand the market for the service and increase the revenue per unit of available capacity"1. Yield management is mostly about maximizing the short term gain of an enterprise. One has to monitor price offers of competitors and has to reflect various constraints over offering products and prices in various booking channels. • Brand Management "is the application of marketing techniques to a specific product, product line, or brand"2. Here, the long term value of a company is the focus of interest. A brand has been carefully monitored, has an active dissemination strategy and can furthermore be used to maintain and increase the value of the brand. Finally crisis management, i.e., rapid cycles of monitoring and disseminating, has to be performed to protect against significant value loss of a brand. • Reputation Management "also known as directory management, is the process of tracking an entity's actions and other entities' opinions about those actions; reporting on those actions and opinions; and reacting to that report creating a feedback loop" 3. It can be the general reputation of an enterprise, as well as a public body, political parties etc. Here it is not directly about an economic value, however, about maximizing the impact of a campaign in relation to a certain budget.

CVM addresses these challenges especially in regard to the increasing number of on-line communication channels such as:

• Broadcasting, e.g.: web sites, news, email, RSS feeds, Twitter, chats, blogs • Sharing of information items such as: bookmarks, images, slides, and videos • Collaboration through wikis

                                                                                                               1 The Basics of Revenue Management, Integrated Decisions and Systems, Inc., 2005. http://www.adhp.org/pdf/1-

theBasicsofRM.pdf 2 http://en.wikipedia.org/wiki/Brand_management 3 http://en.wikipedia.org/wiki/Reputation_management

  4  

• Group communication and interaction through sites such as Facebook, Google+, LinkedIn, XING, and a multitude of specific recommendation and feedback places for certain vertical domains.

• Semantic-based Dissemination through formats such as microdata, RDF, RDFa, and SPARQL; and vocabularies such as Dublin Core, FOAF, and GoodRelations, schema.org, etc.

We want to emphasize the aspect of bi-directionality in these interactions. This means multichannel dissemination, market and feedback analysis of various communication events, and communication in general as integrated loops of both activities.

The 1st International Workshop on Common Value Management focuses on the scientific and technical aspects of semantics in relation to this topic. In this context, the aims of the workshop are to get an insight in the current state of the art in this field and also to mark current academic and industrial trends in this field. Semantics can hereby be used to describe and establish content, channels, and their alignments. Five invited talks, given by Rainer Babiel, Florian Engel and Roman Zimmerman, Martin Hepp, Yannis Charalabidis and Julius van de Laar provide the context for much of the work presented in the workshop papers. The very enlightening and engaging invited talks cover aspects ranging from use of semantics in eCampaigning, exposing business information using semantic technologies, new ways to generate value from and for society with ICT as well as how to shape public perception, frame the debate and effectively engage the audience in the digital age.

Four high quality papers were accepted at the workshop. The paper “Social Media Matrix Matching Corporate Goals with External Social Media Activities” proposes a matrix based instrument that supports companies to decide based on their corporate or communication goals which social media activities to execute. The matrix is   evaluated on reference case social media consultancy projects. The paper “A vision of effective, efficient multi channel online value management” proposes an approach for scalable, effective and efficient online communication using semantic technologies for common value management. The paper “Diversity through Social Media” describes an approach for interactive diversity analysis based on social media that can be used for tasks such as reputation management, market analysis or sentiment analysis. The proposal approach is evaluated on a dataset of social media microposts from telecommunication domain. Finally, the paper “An approach for evaluation of social media monitoring tools.” proposes a structured evaluation framework comprising a set of evaluation criteria that can be used to analyze social monitoring tools from three perspectives: the concepts they implement, the technologies used and the user interface they provide.  

May 2012 Dieter Fensel

Holger Kett

Marko Grobelnik

  5  

Table of Contents

 1      INVITED  TALKS...........................................................................................................................6  1.1  SOCIAL  MEDIA  MONITORING  -­‐  TODAY'S  LEAK  OF  SEMANTICS   .......................................................... 6  1.2  ON  NEW  WAYS  TO  GENERATE  VALUE  "FROM  AND  FOR"  SOCIETY  WITH  ICT................................... 6  1.3  SEMANTIC  TECHNOLOGIES  AND  ECAMPAINING .................................................................................... 7  1.4  TALKING  TO  THE  MARKET  WITH  DATA:    FROM  SEMANTICS    SEO  TO  LINKED  OPEN  COMMERCE ............................................................................. 8  1.5  CUTTING  THROUGH  THE  NOISE:  HOW  TO  SHAPE  PUBLIC  PERCEPTION,  FRAME  THE  DEBATE  AND  EFFECTIVELY  ENGAGE  YOUR  AUDIENCE  IN  THE  DIGITAL  AGE .................................................................... 8  

2.  RESEARCH  PAPERS.....................................................................................................................9  2.1   SOCIAL  MEDIA  MATRIX    MATCHING  CORPORATE  GOALS  WITH  EXTERNAL  SOCIAL  MEDIA  ACTIVITIES.    HARRIET  KASPER,  IVA  KOLEVA,  AND  HOLGER  KETT .............................................................................................. 9  2.2   COMMON  VALUE  MANAGEMENT  -­‐  BASED  ON  EFFECTIVE  AND  EFFICIENT  ON-­‐LINED  COMMUNICATION.    DIETER  FENSEL,  BIRGIT  LEITER,  STEFAN  THALER,  AND  ADNREAS  THALHAMMER ........ 19  2.3   DIVERSITY  THROUGH  SOCIAL  MEDIA.    TAEDJ  ŠTAJNER,  AND  INNA  NOVALIJIA ................................................................................................................ 41  2.4   AN  APPROACH  FOR  EVALUATION  OF  SOCIAL  MEDIA  MONITORING  TOOLS.    IOANNIS  STAVRAKANTONAKIS,  ANDREEA-­ELENA  GAGIU,  HARRIET  KASPER,  IOAN  TOMA,  AND  ANDREAS  THALHAMMER..................................................................................................................................................... 52  

 

  6  

Social Media Monitoring – today’s leak of semantics

Rainer Babiel1

1 Babiel GmbH, Germany www.babiel.com

Abstract: Social Media activities are based on the results of monitoring hundreds or thousands of websites, blogs and social networks. Regarding to the themes which are monitored, daily millions of documents have to be checked and the relevance for the monitoring has to be weight. Reducing the workload to a few thousand articles, photos and so on, the Monitoring has to decide which of them are worth to be read, answered or rectified. All is done on statistical methods without semantic interpretation. And this leads to misinterpretation and fail decisions.

On new ways to generate value “from and for” society with ICT

Yannis Charalabidis1

1 University of Aegean yannisx@aegean.gr

Abstract: Information and Communication Technologies have an unprecedented potential to improve the responsiveness of governments to the needs of citizens and have long been recognized as a key strategic tool to enable reforms in the public sector. During the last years, this potential of ICT has started to be dealt with in a multi-disciplinary way, giving birth to new research communities dealing with governance and policy modeling, modeling and simulation of complex systems, public administration information systems, open governance and social media. During the last 30 years, ICT-related modernization efforts in the Public Sector can be categorized under two main areas: the development and deployment of electronic service portals and systems that targeted the one-stop service provision towards citizens and businesses (wave 1) and the introduction of collaborative tools and social media platforms external to the administration (wave 2), aiming at enhancing the democratic debate without through a clear path from deliberation to implementation of public policies. A third wave in now rising, that will target the heart of governance – that is citizen and small business participation in all aspects of governance, including knowledge management, policy agenda setting, decision making and control. This effort is crucial if large societal problems are to be attached, within an environment of resource scarcity and need for public sector streamlining. Key technologies and scientific domains that will give rise to this new wave of innovative, multidisciplinary and, eventually, breakthrough applications are the following: • Crowd-sourcing methods and tools for getting ideas, development effort, as well as

data by large numbers of citizens

  7  

• Social innovation and revolutionary forms of policy making and governance • Internet activism, acting as a fore-front and ice-breaker in breakthrough ideas • Model-based governance, policy modeling, simulation and visualization • Opinion mining, sentiment analysis and visualization of public deliberation • Trend and behavior forecasting • Open data gathering, curation, publishing and automated access • Open services, offered as basic ingredients for citizen-centric development • Web of Things applications in public administration • Cross-country, cross-language, cross-culture applications for governance

This new wave of ICT-enabled governance applications will eventually assist in tackling large societal problems, such as hunger, poverty, unemployment, immigration, financial crisis, ageing. This large-scale attempts are envisaged to require innovative multi-disciplinary approaches, gathering patterns and solution elements from social and political sciences, mathematics, management, engineering, complex systems theory as well as information and communication technologies. Finally, this third wave for governance applications will also signal the shifting of focus from the “well known stakeholders” that is political parties, administration, members of parliaments, government officials and the sort, towards the “real owners” of the problems and beneficiaries of the final solutions, that is non-governmental organizations, associations of dynamic very small enterprises, student and youth communities, collective citizen bodies of any sort and individuals.

Semantic technology and eCampaining

Florian Engel1 and Roman Zimmerman1

1 more onion, Austria www.more-onion.com

Abstract: Semantic technology can be used for digital campaigns in a number of different ways. One of the areas ecampaigners spend a log of time with is profiling their audience. To reach optimal participation and engagement it's a good practice to test your campaigns, iterate and keep testing. In some ways RDFa and sentiment detection can be used to support this process. Data is key to understanding what's really going on. In most cases there are more questions than answers. Why does one campaign work while the other one doesn't really attract people's attention? Which supporters respond to what kind of content? How can we improve overall engagement of specific supporter groups?

  8  

Talking to the Market with Data: From Semantic SEO to Linked Open Commerce

Martin Hepp1

1 Hepp Research GmbH, Germany www.heppnetz.de

Abstract: In this talk, we will analyze why exposing business information using semantic technology is a key component of any marketing strategy for specific products, and illustrate the roles of schema.org, GoodRelations, RDFa, and Microdata for future Web architectures that reach out to the digital devices of potential customers.

Cutting through the noise: How to shape public perception, frame the debate and effectively engage your audience in the digital age

Julius van de Laar1

1 Strategic Campaigns & Communication, Germany www.juliusvandelaar.com

Abstract: Successful political campaigns have mastered the tactics and strategies used to effectively present an argument, manage and respond with authority during crisis, influence the debate and shape public perception. Yet, in today’s 24/7 media environment it has become more difficult than ever to set an agenda, frame an issue or engage an audience. Four years ago, Barack Obama set a new standard for campaigning by changing the way new media was used to build an aspirational brand, engage and empower supporters, raise money and turn out voters. As the 2012 presidential race unfolds, the campaigns are stepping up their game. And in this cycle, they are embracing digital media more than ever. However, it’s not only the President’s campaign and his opponents who are faced with the challenge to create a narrative and frame the public debate. Organizations in the private sector often deal with the similar complex issues as they struggle to deliver tailored messages to target their audience, regardless of whether its costumers, investors, media, the general public or even potential employees.

From storytelling to big data lifestyle targeting: Julius van de Laar will provide a first hand account on how today’s most effective campaigns leverage battle tested strategies combined with new media tools to create a persuasive narrative and how they translate into actionable strategies for the corporate context.

  9  

Social Media Matrix

Matching Corporate Goals with External Social Media Activities

Harriet Kasper1, Iva Koleva and Holger Kett1

1 Fraunhofer Institute for Industrial Engineering IAO, 70569 Stuttgart, Germany {harriet.kasper, holger.kett}@iao.fraunhofer.de

kolevaiva@yahoo.de

Abstract. In this paper we introduce the Social Media Matrix, a practitioner-oriented instrument that supports companies to decide based on their corporate or communication goals which social media activities to execute. The matrix consists of three parts: 1. Social media goals and task areas have been identified and matched. 2. Five types of social media activities have been defined. 3. The matrix provides a structure to assess the suitability of each activity type on each social media platform for each goal. Whereas the first two parts can be generally used, the assessment must be conducted explicitly for an industry sector. A ready to use assessment for the German B2B sector has been exemplarily compiled from expert-interviews with practitioners and by reviewing concrete social media activities. The matrix is used as a basis for social media consultancy projects and evaluated thereby.

Keywords: web 2.0, social media, corporate social media, marketing, communication, planning, instrument, tool, decision tool, practitioner-oriented, goal-oriented, business-to-business, B2B.

1 Introduction

Facebook, founded in 2004, according to its own website4 had 845 million monthly users at the end of December 2011. Together with platforms like Twitter5, LinkedIn6 or YouTube7 and other web 2.0 sites like blogs and forums, social media’s impact on society and its relevance for business cannot be denied. The Social Media Governance Study 2011 [1], an empirical survey of communication managers and PR professionals in companies, governmental institutions and non-profit organizations in the German-speaking part of Europe: Germany, Austria and Switzerland, reveals that 71.4 percent of all organizations currently actively apply social media in their communications, which is a growth of 17 percent compared to the survey result of 2010 [2]. Only 7 percent of the respondents have neither applied nor planned to use social media. In the current project CLOUDwerker8, which is partially funded by the German Ministry of Economy and Technology, we experienced in several interviews that even small crafts enterprises are aware of the importance of social media and want to enter this field. Furthermore a survey of the University of Darmstadt [3] examined social media use by business-to-business (B2B) oriented companies, which seems to be different from consumer oriented businesses (B2C) and therefore is also a topic we are addressing with this paper. Structured approaches, methods and tools that support the process of choosing the `right` social media activities for a company have for a wide range of industries and businesses not been introduced yet. Our work aims to close this gap and suggests an instrument that can provide the necessary backing for corporate social media decisions. In this paper we concentrate on external social media activities, in contrast to such activities that are dealing with the internal use of social media also referred to as enterprise 2.0 [4].                                                                                                                4 www.facebook.com 5 www.twitter.com 6 www.linkedin.com 7 www.youtube.com 8 www.cloudwerker.de

  10  

Current literature concentrates on different aspects of social media and is examined in section 3 of our paper, after introducing our methodology for developing the Social Media Matrix in section 2. In section 4 we will present the overall structure of the Social Media Matrix. Section 5 discusses the exemplary assessment of social media activities. We show how the Social Media Matrix can be used in section 6 and conclude with describing future extensions in section 7.  

2 Methodology

The basic idea of matching corporate and communication goals with concrete social media activities on social media platforms arouse from talking to practitioners in companies across different industry sectors. Although it was our intention to create an instrument which is widely applicable, in the initial phase of our work we have chosen to narrow down our target group on business-to-business (B2B) oriented companies in Germany. Besides considering relevant literature we have carried out six structured expert interviews [5] with practitioners who are in charge of social media activities in such B2B companies to identify their:

• overall approach, perceived potentials and challenges • goals that should be met by social media activities • best practices from experience and observation

The evaluation of the documented interviews confirmed the necessity of a structured approach, served as a review and completion of our goal list and formed the base for the assessment presented in section 5. Furthermore by including the intended end user of the Social Media Matrix its relevance to practice could be ensured. Our assumption that it is essential to assess social media activities and accumulate best practices and examples for an industry sector rather than only for goals has confirmed and today during the evaluation in consultancy projects our matrix can easily be adapted to other sectors e.g. business-to-consumer oriented industries.

3 Literature Review

Concerning social media, companies must answer the questions what to do (strategy), how to do this (organization: roles and processes) and which tools offer support to do so (information systems). This approach can be derived from Oesterle [6] who suggested this classification for business engineering tasks in general. Furthermore our practitioner-oriented view asks to distinguish between three solution types:

• structured approach for choosing and optimizing concrete social media activities • models, methods and checklists for designing and implementing social media activities • instruments supporting structured approach, including models, methods and checklists

Figure 1 gives an overview of social media publications and classifies them according to the above mentioned criteria.

  11  

Table 1. Classification of social media literature

Additional aspects of our literature review are summarized in the following paragraphs:

• Scientific level: A general distinction of literature about social media can be drawn between scientific studies e.g. Kaplan and Haenlein [7], Mangold and Faulds [8], Hettler [9], Peters [10], Zerfaß [11] and a tremendous number of key expert views on the use of social media in business e.g. Weinberg [12], Brogan [13-14], Evans [15], Li and Bernoff [16] and Solis [17]. Due to being a fast moving new field of study, many of the experts and key authors in social media are current practitioners in this field, who have been widely-accepted as thought leaders e.g. key bloggers like Jeff Jarvis9.

• Definition of social media: There is a discussion about the concept of social media and how it differs from related concepts such as web 2.0 and user generated content [7][12][18-20]. The authors provide general advices for companies on how to engage in social media and how to communicate with their stakeholders on different platforms. In this regard Joel [19] contends that marketing of trust and transparency is important and focuses on the idea of ‘building of a community based on trust’. Evans [15] shares this point of view and emphasizes the importance of involving customers in social media conversations with regard to social customer relationship management.

• Impact of web-based activities: Many early approaches consider not only technological development but also the changes in the ways organizations and end-users use the web. As one of the first books on web 2.0 ‘The Cluetrain Manifesto’ [21] provides a set of 95 theses, which aim to examine the impact of the internet on both consumers and organizations. The manifesto contends that internet technologies enable people to have "human to human" conversations, which have the potential to transform traditional business practices. O’Reilly [18] describes the revolution of Web 2.0 as the era of participation and collective intelligence, and with his book “The wisdom of crowds” [22] James Surowiecki provided a standard work on the chances of this new development beyond using it for PR and marketing purposes.

• Structured approach: Few studies [9-11], [16] provide structured approaches, suggesting strategies and models for social media engagement and for implementation of social media

                                                                                                               9 www.buzzmachine.com/about-me/

  12  

activities. Forrester Research executives Li and Bernoff [16] develop a framework for engaging in the ‘groundswell’ based on following activities: listening, talking, energizing, and supporting. The framework attempts to help companies to understand and engage with their customers within the social media space. Although Li and Bernoff suggest a clear community engagement model and segment web 2.0 participators, it does not satisfactorily provide an understanding of the contribution of these activities to corporate objectives. Zerfaß [11] suggests the concept of ‘Social Media Governance’ as a regulatory framework for incorporating social media, based on a quantitative survey in Germany [1-2].

• Overall social media concept: Although ‘Social Media Governance’ by Zerfaß is an overall concept concentrating on organizational issues, it focuses on the PR-perspective and omits that social media can be used to achieve goals in fields such as product management or recruiting. Other studies analyze the use of social media from specific viewpoints, for example PR [23], social media marketing [12][15], reputation management [10].

• Social media marketing: Some authors [12], [24] concentrate on social media marketing and understand social media as a chance to promote websites, products or services and ‘to communicate with and tap into a much larger community that may not have been available via traditional advertising channels’ [12]. Others are adamant that social media and marketing are distinctive disciplines and independent from each other. Brogan [14] for example suggests that “Marketing is NOT Social Media - Social Media is NOT Marketing”. Focusing on the PR-perspective, Scott [23] believes that traditional public relations practices ‘do not work anymore’. He contends that web 2.0 allows companies to take ownership of information and independently publish it instead of pushing it in the form of press releases to traditional media outlets. Although a lot of works analyze social media from a specific perspective, only a few studies focus on the integration of social media in the business practice. Mangold and Faulds [8] see social media as an element of the promotional mix within the marketing mix of a company and stress the need for integration of social media activities into communication strategies. They provide a set of examples for companies, which have successfully integrated social media. However, the study does not contain a well-structured approach, explaining how social media can be incorporated.

• Social media management tools: For monitoring and analysis of social media activities a new tool class has established which plays an important role in social media management. The provided opportunities for companies using these tools and their specific functions have been examined in several market studies [25-27]. The publications see a trend of these tools developing towards engagement platforms, which allow companies to directly react on found contributions in social media and a necessity to make the functions available for workgroups. In a whitepaper Owyang and Lovett suggest a social media measurement framework [28] and projects like Next Corporate Communications10 research value and return on invest (ROI) of social media, but general, in depth and ready to use instruments and key performance indicators have not yet been revealed.

• Model for social media planning: Despite the high interest in the field of social media, well-structured approaches and models for realizing social media activities in the business practices are to our knowledge limited available. Although some scientific studies as well as some key experts suggest that planning social media activities has to be a part of a corporate communications strategy [8-9], [11], still they do not satisfactorily provide solutions that help companies to plan and coordinate activities.

As shown in figure 1 current literature focuses on ‘strategy’ but when it comes to the more practice-oriented views on ‘processes’, ‘information systems’ and concrete ‘models, methods and checklists’ most generalizations do not work anymore. With the Social Media Matrix we suggest an ‘instrument supporting structured approach, including models, methods and checklists’.

                                                                                                               10 http://www.nextcc.de/

  13  

4 Social Media Matrix and its overall structure

Our work supports the methodical selection of appropriate social media activities. The suggested instrument ‘Social Media Matrix’ aims at providing a match of suitable social media activities to relevant social media goals of companies. Thus, the matrix provides (1) a classification of social media goals and dedicated business tasks, (2) a categorization of social media activity types, and (3) the concrete assessment of each activity type on each platform for each goal, including a collection of examples. Figure 2 illustrates the overall structure of the Social Media Matrix.

Table 2. Structure of the social media matrix

(1) Classification of relevant social media goals by task areas

Social media can be used for achieving various goals of different departments. The matrix provides a classification of social media goals by task areas, in order to allow the user to focus on analysis, planning, execution and controlling of social media - away from the company’s functional perspective and toward the integrated communication management concept [8], [11]. To facilitate orientation and coordination with regard to the practical application of the instrument, we have identified different social media task areas and matched these with the goals. The following eleven task areas have been assigned to a total of 29 social media goals:

• media relations • reputation management • agenda setting • issues management • crises management • branding & brand management • customer relationship management & influencer relationship management • trend, market & competition analysis • social media marketing & social commerce • market research & monitoring

  14  

• product & innovation management

A task area may pursue more than one goal. Furthermore the single goals may impact each other. For this reason, we have distinguished between main and secondary goals within a task area. For example, ‘increasing brand awareness’ is defined as a main goal within the task area ‘branding & brand management’. Whereas ‘building relationships to influencers in social media’ is a secondary goal of ‘branding and brand management’ which will contribute to the achievement of the main goal.

(2) Categorizing activities

A major outcome of our work was to realize that before assessing the suitability of a platform for a goal the type of activity must be distinguished. We have therefore defined five types of social media activities, which cover all possibilities:

• content: Content refers to the publishing of information in the form of text, image, graphic, audio and video. This category includes only activities, which are based on one way communication, e.g. a company’s info on the Facebook fan page or the channel on YouTube.

• interaction and dialog: This type comprises activities aimed at interacting and discussing with stakeholders. Such activities can be for example posts on the company’s wall on Facebook or a group on LinkedIn.

• listening and analyzing: This activity type is about finding conversations about the company, its brands and products and analyzing these in order to turn them into valuable insights. This monitoring can be manual or automated and can furthermore include keyword search about competitors, current issues, important stakeholders etc.

• application: Expanding the possibilities of a social media platform by own applications to enable further interaction is summarized under this activity type. For example Facebook allows programming of own applications.

• networking: Taking active steps to find contacts and to build relationships is referred to as networking. This includes e.g. @-referencing other users on Twitter or inviting users to a group on LinkedIn.

(3) Assessing platforms and finding examples

Whereas goals, task areas and activity types are generally applicable this third part of the Social Media Matrix must be adapted both to the industry sector and the country/regional context. It comprises a list of the most relevant social media platforms in one country and the assessment of the suitability of each activity type on each platform for each goal. Additionally best practices are collected and added as examples to the assessment. The assessment can either rely on expert rating or derive from examples and best practices. For our exemplary assessment for B2B companies in Germany we used a combination of both approaches. A third possibility which we shortly discuss in the ‘future research’ section is crowdsourcing the assessment.

5 Assessing suitability of social media activities for German B2B companies

To make the Social Media Matrix a ready to use instrument for planning corporate social media activities it must include specific assessment and examples to answer the question “How useful is a certain activity type on a certain platform to reach a certain goal?”. In contrast to social media goals and to a certain extend also platforms which can both be generalized, the assessment in the matrix must be explicit, e.g. for an industry sector. In our work social media activities of German companies that predominantly follow business to business (B2B) transactions have been reviewed and evaluated according to goal, activity type and used platform to sample such a specific assessment. Together with the results of the

  15  

expert interviews a directly applicable Social Media Matrix for the German B2B sector has been created which is at the same time through the examples a snap-shot of current social media use in B2B companies in Germany.

The social media platforms we considered in this use case were Facebook, Twitter, YouTube, Xing11, Blogs and ‘community platforms or forums’. For transferring the results to other than German markets the platform Xing should be substituted by LinkedIn which is its international pendant. For most goals content, interaction / dialog and networking are the predominant activity types. Our assessment shows that social media activities can be used in German B2B companies best to achieve these goals:

• building and improving product and brand image • identifying, contacting and bonding influencers • improving ranking on search engines • generating leads • employer branding and recruiting

Social Media activities that aim towards vending products or providing service and support over social media platforms are less suitable, due to the complexity of B2B products. Nevertheless it could be a strategy to explicitly address such goals that seem less suitable and are therefore less represented to create unexpected social media activities that by their newness gain more attention which in social media means success.

6 Application Modes

Integrated communication distinguishes four management phases: analysis, planning, execution and controlling [29]. The Social Media Matrix is primarily designed to support planning activities, but can also be used for analyzing industry sectors beforehand or controlling own activities. A second dimension in applying the Social Media Matrix is the starting point of decision making. Besides goals - platforms and best practices can be used. The proposed structure allows all three starting points. Furthermore target groups / stakeholders are often used as a starting point for activity analysis, planning and controlling. This perspective has not (yet) been considered in the matrix, but since target groups are an implicit part of goals and best practices further detailing the matrix in this respect is possible.

                                                                                                               11 www.xing.com

  16  

Figure 3 illustrates the application modes of the Social Media Matrix.

Fig. 1. Application modes of the Social Media Matrix

We have initially implemented the social media matrix in Excel. The user can flexibly use the matrix for decision making by for example:

• prioritizing goals or filtering goals by task area • using the assessment of the platforms to decide upon which platform to enter or to

understand which optimization opportunities already used platforms offer • using the assessment to identify which goals can best be achieved by social media activities • using the given examples to better understand possibilities

7 Conclusion and future research

We have introduced an instrument for companies to choose social media activities based on corporate goals and an assessment of the suitability of certain activities on selected platforms. The suitability assessment has been conducted exemplary for the German B2B-sector. Moreover our Social Media Matrix contains an example set of realized social media activities in the B2B-sector in Germany, which can be regarded as best practices and provide orientation for the user. Therefore this is a ready to use tool for German B2B-companies which is currently evaluated during consulting projects. It has been shown, that our practical integrated approach has not been presented in literature so far. This work builds the foundation of a more sophisticated tool for planning and controlling corporate social media activities. To close this paper, three selected aspects of future work are presented in brief in the following paragraphs.

• Crowdsourcing example collection and assessment of social media activities: Especially in the field of social media the web community is a great source for information which can be used for optimizing the Social Media Matrix. A web interface could be used to collect examples for corporate social media activities. Each example should be assigned at least one goal, the industry-sector, the platform and the activity type it refers to. In an international context, country information should also be provided. Incentive to enter this data could be the possibility to compare the entered activity with other activities in the same industry sector. Assessing activities for different sectors could be either by experts, or everybody. Exceeding critical masses is essential here.

• Detailing social media platform representation: Since marketing and corporate communications is often driven by target group segmentation, this information must be modeled within the social media matrix. One approach is to add target group information to

  17  

the description of platforms. Ideally the defined parameters can be easily used to describe any platform and new platforms thereby could derive their assessment from existing data.

• Estimating costs: To make the Social Media Matrix an effective tool in terms of Common Value Management the assessment of social media activities must be expanded by adding cost factors for execution of an activity. Cost factors could be distinguished into initial cost and operating cost, they contain time elements (e.g. singular vs. regular) and are interconnected with the available resources. It is a matter of research to what extend cost estimate and ROI of social media activities can be generalized.

8 Acknowledgements

The work published in this article is partially based on the diploma thesis of Iva Koleva filed in January 2012 with the University of Hohenheim in Stuttgart, Germany. The thesis has been initiated and supervised by Harriet Kasper, Fraunhofer IAO and examined by Prof. Dr. Frank Brettschneider-Vetter (1st examiner, University of Hohenheim) and Prof. Dr.-Ing. Dr.-Ing. E.h. Dieter Spath (2nd examiner, University of Stuttgart).

References

[1] Fink, S., Zerfaß, A., Linke, A.: Social Media Governance 2011 – Expertise Levels, Structures and Strategies of Companies, Governmental Institutions and Non-Profit Organizations communicating on the Social Web. University of Leizpig / Fink & Fuchs Public Relations AG, Leipzig / Wiesbaden (2011). Retrieved from www.socialmediagovernance.eu

[2] Fink, S., Zerfaß, A.: Social Media Governance 2010 – How companies, the public sector and NGOs handle the challenges of transparent communication on the Internet. University of Leipzig / Fink & Fuchs Public Relations AG, Leipzig / Wiesbaden (2010). Retrieved from www.socialmediagovernance.eu

[3] Pleil, T.: Mehr Wert schaffen: Social Media in der B2B-Kommunikation. Books on Demand, Darmstadt (2010)

[4] Spath, D. (ed.), Guenther, J.: Trendstudie Wissensmanagement 2.0 – Erfolgsfaktoren für das Wissensmanagement mit Social Software. Fraunhofer Verlag, Stuttgart (2010)

[5] Bogner, A., Littig, B., Menz, W.: Das Experteninterview – Theorie Methode, Anwendung. 2nd ed., Wiesbaden: VS Verlag für Sozialwissenschaften (2005)

[6] Oesterle, H.: Business Engineering – Prozess- und Systementwicklung. Band 1: Entwurfstechniken. 2nd ed., Springer-Verlag, Berlin (1995)

[7] Kaplan, A. M., Haenlein, M.: Users of the world, unite! The challenges and opportunities of social media. In: Business Horizons, vol. 53 (1), pp. 59 - 68 (2010)

[8] Mangold, W. G., Faulds, D. J.: Social media - The new hybrid element of the promotion mix. In: Business Horizons, vol. 52 (4), pp. 357-365 (2009)

[9] Hettler, U.: Social Media Marketing. Marketing mit Blogs, sozialen Netzwerken und weiteren Anwendungen des Web 2.0. Oldenbourg, München (2010)

[10] Peters, P.: Reputationsmanagement im Social Web. Risiken und Chancen von Social Media für Unternehmen, Reputation und Kommunikation. Social Media Verlag, Köln (2011)

[11] Zerfaß, A., Fink, S., Linke, A.: Social Media Governance: Regulatory frameworks as drivers of success in online communications. IPRRC International Public Relations Research Conference, Miami (2011)

[12] Weinberg, T.: Social media marketing. Strategien für Twitter, Facebook & Co. O'Reilly, Beijing (2010) [13] Brogan, C., Smith, J.: Trust economies: investigations into the new ROI on the web. (2008) Retrieved

from http://changethis.com/44.04.TrustEconomy [14] Brogan, C.: Marketing is NOT Social Media-Social Media is NOT marketing. (2007) Retrieved from

http://www.chrisbrogan.com/marketing-is-not-social-media-social-media-is-not-marketing/ [15] Evans, D.: Social media marketing. An hour a day. Wiley, Indianapolis, Indiana (2008) [16] Li, C., Bernoff, J.: Groundswell: winning in a world transformed by social technologies. Harvard Business

School Press, Boston, MA (2008) [17] Solis, B.: Engage! The complete Guide for Brands and Business to Build, Cultivate and Measure Success

in the New Web. Wiley, Hoboken, N. J. (2010) [18] O'Reilly, T.: What Is Web 2.0. O'Reilly Media (2005). Retrieved from

http://oreilly.com/web2/archive/what-is-web-20.html [19] Joel, M.: Trust economies: the new marketing ROI. Six Pixels of Separation (2007). Retrieved from

http://www.twistimage.com/blog/archives/trust-economies---the-new-marketing-roi/

  18  

[20] Joel, M.: Six pixels of separation. Everyone is connected - connect your business to everyone. Business Plus, New York (2010)

[21] Levine, R., Locke, C., Searls, D., Weinberger, D.: The cluetrain manifesto. Perseus Publishing. http://www.cluetrain.com/ (2000)

[22] Surowiecki, J.: The Wisdom of Crowds - Why the Many Are Smarter Than the Few and How Collective Wisdom Shapes Business, Economies, Societies and Nations. Doubleday, Anchor (2004)

[23] Scott, D. M.: The new rules of marketing & PR - How to use news releases, blogs, podcasting, viral marketing and online media to reach buyers directly. Wiley, Hoboken, N. J. (2008)

[24] Barefoot, D., Szabo, J.: Friends with benefits. A social media marketing handbook. No Starch Press, San Francisco (2010)

[25] Kasper, H., Dausinger, M., Kett, H., Renner, T.: Marktstudie Social Media Monitoring Tools – IT-Lösungen zur Beobachtung und Analyse unternehmensstrategisch relevanter Informationen im Internet. Fraunhofer Verlag, Stuttgart (2010)

[26] Plum, A.: Ansätze, Methoden und Technologien des Web-Monitorings – ein systematischer Vergleich. In: Brauckmann, P. (ed.): Web-Monitoring – Gewinnung und Analyse von Daten über das Kommunikationsverhalten im Internet. Pp. 21-46, UVK, Konstanz (2010)

[27] Gilliatt, N.: Social Media Analysis Platforms for Workgroups. Social Target LCC (2010) [28] Owyang, J., Lovett, B.: Social Marketing Analysis – A New Framework for Measuring Results in Social

Media. Altimeter Group / Wen Analytics Demystified (2010). Retrieved from http://www.slideshare.net/jeremiah_owyang/altimeter-report-social-marketing-analytics

[29] Zerfaß, A.: Unternehmenskommunikation und Kommunikationsmanagement - Grundlagen, Wertschöpfung, Integration. In: Pilwinger, M., Zerfaß, A. (eds.): Handbuch Unternehmenskommunikation. Pp: 21-70, Gabler, Wiesbaden (2007).

  19  

Common Value Management -

based on Effective and Efficient On-line Communication

Dieter Fensel1, Birgit Leiter1, Stefan Thaler1, and Andreas Thalhammer1

1 University of Innsbruck, STI-Innsbruck,

6020 Innsbruck, Austria firstname.lastname@sti2.at

Abstract. We discuss the challenge of scalable dissemination and communication approaches

in a world where the number of channels is growing exponentially. The web, Web 2.0, and semantic channels have provided a multitude of interaction possibilities providing significant potential for yield, brand, and general reputation management. Our goal is to enable smaller organizations to fully exploit this potential. To achieve this, we have developed a new methodology based on distinguishing and explicitly interweaving content and communication as a central means for achieving content reusability, and thereby scalability over various heterogeneous channels.

Keywords: brand-, reputation-, yield- and value management

1 Introduction and Motivation

The last two hundred years have revolutionized international transport and communication. Fax, phone, and later the Internet, have radically changed our communication possibilities. More and more communication has been freed from the geographic barriers that formerly limited their speed and expansion. Now, it is (in principle) possible to instantly communicate with a large portion of the entire human population. Nevertheless, new means also generate new challenges. Take the world of the TV consumer as an example. Twenty-five years ago, there were around three channels. Therefore, selecting your program was a rather trivial task which required no more than a few seconds. Whilst hundreds of channels have been added, thousands of channels have been connected via the Internet, where extremely large libraries of videos (which go beyond the metaphor of a ‘channel’), currently define the content. The consumer could now spend a lifetime in search of a program he or she wishes to watch. Obviously, consumers require new skills and more efficient access means to scale and filter the exponentially increased offer. Precisely the same is needed for our overall approach to on-line (or Internet-based) communication. Assume the task of a small hotelier. How can it be ensured that the hotel is found by potential customers, i.e., how can he/she find them? The hotelier should have a website with high visibility on various search engines and must be present in a large number of on-line booking channels. We should find the hotel on the town's website, and the hotel should have a Facebook page, perhaps with a booking engine included. Bookings made through mobile platforms are increasingly popular, and the hotelier would want to be found there too. Why not add a video about the hotel on YouTube, a chat channel for instant

  20  

communication, fast email and fax response capabilities, the old-fashioned telephone, and occasional tweets and emails that are clearly distinguishable from spam? Preferably the communication should be multi-directional, i.e., the hotelier should realize when one of his posts gets commented on (up to a full-fledged impact analysis), or even more importantly, the hotelier should know when someone talks about the hotel, and how much the costumer liked it. As much as this is needed, this obviously does not scale and [2] calls it “the growth of the multichannel monster”. In principle, the hotelier is presented with three equally problematic alternatives: • He does it on an ad-hoc basis by himself, as a side activity. This would work, however,

the number of potential customers (and therefore business opportunities) that he is missing may be tremendous and could, in the long run, take him out of the market completely.

• He builds up a professional communication team by hiring a large number of social media experts and assigns them to manage his various communication channels. Actually, a large hotel chain may be able to do this. In the case of our hotelier, he would find himself even more quickly out of the market due to the high costs attached to this “solution”.

• Finally, he could start to cooperate with an external marketing agency. This marketing agency must understand the domain (tourism, accommodations) and the various communications means available to disseminate the contents about our hotel in an effective and efficient fashion. These agencies have some IT support that supports multi-channel dissemination, however, they have to manually adopt, align, and define the content for these channels. In summation, these services are costly and only partial solutions (to limit the high costs of manual labor by dissemination experts).

Organizations of all sizes, commercial and not-for-profit, regularly face the challenge of communicating with their stakeholders using a multiplicity of channels, e.g. websites, videos, PR activities, events, email, forums, online presentations, social media, mobile applications, and recently structured data. The social media revolution has made this job much more complicated, because: • the number of channels has grown exponentially, • the communication has changed from a mostly unilateral "push" mode (one speaker,

many listeners) to an increasingly fully bilateral communication, where individual stakeholders (e.g. customers) expect one-to-one communication with the organization, and the expected speed of reaction is shrunk to almost real-time, and

• the contents of communication becomes more and more granular and increasingly dependent on the identity of the receiver and the context of the communication.

Organizations need an integrated solution that provides management and execution of communication goals in a mostly automated fashion, with costs equivalent to mass-media communication, along with the granularity of individual experts, and at the pace of real-time social media. We are aiming to mechanize important aspects of these tasks, allowing scalable, cost-sensitive, and effective communication for small-or-medium sized business units and comparable organizations for which information dissemination is essential but resources are significantly limited. Additionally, it may also help intermediaries such as marketing agencies to extend their business scope by increasing the cost-effective ratio. Communication is a means to an end. Section 2 analyses the major goals that may underlie communicative interaction of an organization with a larger audience. We identify yield, brand, and reputation management as three major aspects around which communication may be centered. They can be identified as variations of general value management differing in their short-versus-long-term orientation towards commercial goals, as well as in their overall connection to financial value orientation. Section 3 sketches the major technical elements that we developed to implement a common value management framework. Section 4 discusses some of the related work and directions for future research activities. Finally, conclusions are provided in Section 5.

  21  

2 The Aspects of Value Management

Scalable, multi-channel communication is a difficult challenge. In order to better understand it, we want to clarify the various underlying goals that it should achieve. Agents often connect (directly or indirectly) economic interests with their communication activities. In the following, we discuss different economic contexts for the communication approaches of organizations.

2.1 Yield Management

Yield or revenue management “is an economic discipline appropriate to many service industries in which market segment pricing is combined with statistical analysis to expand the market for the service and increase the revenue per unit of available capacity” [3].12 Short-term increase of income is a valid target for a business entity; however, it is quite tricky to realize in a multichannel world. For example, hotels are confronted with a multitude of on-line booking channels. Hotels should provide their available rooms and rates to most, if not all of these channels to prevent missing their potential customers. For many channels, visibility is achieved through low prices. However, channels also often require price constraints on the price offers of other channels. Some channels generate costs without guaranteeing actual income. Let us discuss some important issues: A hotel currently provides active booking facilities on its hotel website, through booking.com, expedia.com, Facebook, and Google Maps. It wants to increase the overall conversion rate for bookings in all of these channels, and therefore needs answers to the following questions: • How is my price positioned towards the price of my competitors? • What is my reputation in all of these channels? More than 90% of all Internet users are

already reading product reviews and more than 50% have indicated that their purchasing decisions are based primarily upon them.

• Am I adequately represented in all these channels (comments, reviews, etc.)? • The hotel could reduce its price in channel X to maintain visibility and to increase

bookings through this channel. Due to legal constraints, it has to reduce its price in many other channels which leads to reduced revenues from those bookings.

• A hotel needs recommendations for what needs to be done and the support to do it, e.g. possible actions would be to reduce their price by 10% or to include more amenities and supplements, to be more active on Facebook in order to increase social media links and conversations, or if guests complain about the coffee it could therefore be alerted to improve this service.

• A hotel wants to announce rooms through Google Maps. It therefore starts to pay for each click through this search interface. If many clicks fail to lead to a booking, the hotel may begin to lose a significant amount of money.

• The same scenario could occur with static on-line coupons13 that offer a 50% discount through a coupon platform that requests an additional 25% for each coupon. This can easily end in negative revenue and lacks dynamicity. Imagine a “magic” dissemination button for a bar owner that can announce dynamically a happy hour, special offers, interesting news etc. on-the-fly to the right circle of interested public, establishing the bar as a trendy place where a hip crowd is getting together.

Many solutions to yield management are based on complex statistical methods and complex domain assumptions on how variation of the price can influence the number of bookings of a service. However, a multi-directional multi-channel approach must also rely on Swarm intelligence14. Observing in real time the reaction of customers and competitors will be key to achieving on-line marketing. Adopting your offer and your price dynamically in response to the behavior of your (on-line visible) environment will become critical to economic success.                                                                                                                  12 http://en.wikipedia.org/wiki/Yield_management, and Revenue_management 13 http://en.wikipedia.org/wiki/Coupon 14 http://en.wikipedia.org/wiki/Swarm_intelligence

  22  

2.2 Brand Management

Yield management tries to maximize the immediate revenue of an organization. However, communication is also very important in relation to the long-term value of a company. Actually, the reputation of a company can be viewed as one of its most important assets. Proper management, such as managing the value of brands,15 may be essential for its long-term economic success. This may conflict with revenue management. In many cases, it may be useful for short-term income management to reduce the price of the offering, which on the other hand can diminish and undermine the long-term income that is generated through a general price profile indicating quality and exclusivity.

2.3 Reputation Management

The economic impact of proper reputation management is evident when we talk about the reputation of economic entities. However, non-profit organizations also have a need for general reputation management and public campaigns. 16 “Reputation is the opinion (more technically, a social evaluation) of a group of entities toward a person, a group of people, or an organization on a certain criterion. It is an important factor in many fields, such as education, business, online communities or social status.”17 Here, it is not the direct and intermediate economic income that matters. It is rather about maintaining or increasing the appreciation an organization, topic, or certain approach gains in the public eye. However, even a campaign on a public issue has an immediate economic dimension to it: trying to use the available budget for it in the most effective way. Therefore, providing means to increase the effectiveness and efficiency of public campaigns is of high value.18

2.4 Value Management

All of the issues above could be viewed as facets of Value Management, where value is defined as the regard that something is held to deserve, i.e., its importance. Online, multi-channel and bi-directional Value Management is about disseminating19, communicating, and interacting with large, on-line communities to increase the value of a certain entity or issue. The value managed could cover issues such as importance, economic short-term income, or long-term value. [4] identifies the following activities as part of an on-line based value management: Reputation management; Competitive Intelligence, i.e., Competitor Observation; Market Analysis; Influencer Detection; Trend Analysis; Market Analysis; Crisis Management; Issue Management; Campaign Monitoring; Product and Innovation Management; Customer Relationship Management; Risk Management; and Event Detection. Obviously, these activities overlap and share many common elements. It would be interesting to reduce these activities to the set of atomic tasks from which they are composed.

3 The Aspects of Value Management

We start this section by introducing the underlying idea and major structure of our approach. We then discuss our role model, tool support, and finally, we sketch some applications of our approach.

                                                                                                               15 “The American Marketing Association defines a brand as a "name, term, design, symbol, or any other feature

that identifies one seller's good or service as distinct from those of other sellers.” http://en.wikipedia.org/wiki/Brand

16 E.g. http://www.readwriteweb.com/archives/how_to_manage_your_online_reputation.php. 17 http://en.wikipedia.org/wiki/Reputation 18 Also, in the case of political parties, the number of votes they collect can be seen as their “economic” value. 19 “To disseminate (from lat. disseminare „scattering seeds“), in terms of the field of communication, means to

broadcast a message to the public without direct feedback from the audience.” http://en.wikipedia.org/wiki/Dissemination

  23  

3.1 Separating Content and Channel to enable various dimensions of reuse in transactional communication

“I am about to propose the existence of something called the knowledge level, within which knowledge is to be defined.”20

The core idea of our approach is to introduce a layer on top of the various Internet based communication channels that is domain specific and not channel specific.21 So one has: • information models, that define the type of information items in a domain; • a channel model (or communication model), that describes the various channels, the

interaction pattern, and their target groups; • mappings of information items to channels through weavers; and finally, • a library of implemented wrappers for actual channel instances. What is essential is to distinguish the communication or channel model from the conceptual descriptions of the information.22 Our approach requires the creation of a communication model (i.e., an increasingly complete model of channels), and knowledge models for each vertical (such as research projects, research institutes, associations, hotels, restaurants, tourist events, medical doctors, etc.), and finally linking the knowledge model with the communication model through a weaver that weaves concepts with channels. This will not be cheap! However, you pay one price for an entire vertical, i.e. for thousands or up to millions of business units. Even if it would cost the same amount as doing it manually directly on a channel basis for ten of them, it is obviously cheaper if the number of business units is higher than ten. And on the contrary, we think that even if done directly for one business unit, it quickly turns into profit by saving on manual dissemination work through a mechanized communication model. After all, data and information can be expressed at the conceptual level, which the domain expert understands. Mapping of the different communication means (where he is not at all an expert) is done automatically after the first implementation. The difficult dissemination through channels is done automatically through proper channels that are attached to these concepts. Currently, all commercially available solutions are only channel centric and do not provide any built-in support for what needs to be disseminated or where to disseminate what piece. In our approach, a knowledge-model is built and explicitly linked with the channel model. This must be done once for a hotel, and can then be reused for millions of them. That is, we aim for the major elements of reusability:

1. The same information element can be reused for various channels through its channel independent formulation using the information model. 2. The information model is developed as domain ontology for a certain vertical area such as tourist accommodations, gastronomy, medical doctors etc. Therefore, it can be reused for various agents active in the same vertical domain.

These elements of reusability deliver the major contribution to the scalability of our approach.

3.1.1 Information Model

An information model is an ontology that describes the information items that are used in typical communication acts in a certain domain. Many methodologies for building such ontologies have been developed, compare [5]. Building ontologies can be a time-consuming and expensive process. Fortunately, we have a strong modeling bias that helps us to

                                                                                                               20 This notion of knowledge level was first introduced by Allen Newell in the 1980s, to have a way to rationalize

an agent’s behavior. ... Beneath the knowledge level resides the symbol level. Whereas the knowledge level is world oriented, namely that it concerns the environment in which the agent operates, the symbol level is system oriented, in that it includes the mechanisms the agent has available to operate. The knowledge level rationalizes the agent’s behavior, while the symbol level mechanizes the agent’s behavior. http://en.wikipedia.org/wiki/Knowledge_level

21 See also as an excellent presentation on this idea: http://www.slideshare.net/reduxd/beyond-the-polar-bear 22 In analogy to style sheets that separate the contents from its presentation.

  24  

significantly guide and therefore reduce such an effort. We do not aim to model a domain as such, through a very deep model that allows arbitrary (transactional) applications. We can rather focus on the major and typical information items that are used in the on-line dissemination and communication processes. Therefore, the size of these ontologies in our case studies (see Section 3.4.), were moderate (around 100 concepts and properties), and many of these concepts and properties could be reused between different use cases. As a result, there was a reduced effort in building informal domain models (less than one person month). After defining an informal model, we formalized this ontology (see [6] for more details) in a simple sublanguage of OWL-2, since we foresee little need for reasoning about it. We model structured information items as concepts and non-structured ones as properties, i.e., we assume simple non-structured values for properties. Actually, we separated a generic ontology common to all three use cases and specific refinements. The classes of the core ontology are depicted in Figure 2 and its properties in Figure 3.  

 Fig. 2. Classes of the core ontology

 

Fig. 3. Properties of the core ontology

  25  

In general, our models are rather simplistic, but this reflects the fact that we do not model a domain with all its complexity but rather the information chunks that are disseminated about it. That is, our modeling perspective and grainsize are defined by the consideration that an information provider can make informal sense of a chunk and that this chunk can be mapped onto a communication channel. In an intermediate phase of our journey, we also tried to directly use some LOD vocabularies to model these ontologies. 23 However, we took some important lessons from this enterprise:

• Our domain models were suddenly unintelligible to the domain experts. The LOD vocabularies used different terms and made different and non-intuitive modeling decisions compared with our informal domain models (especially not the ones in the eTourim case study).

• It was extremely hard to decide which term to take from which vocabulary. The terminologies were either redundant or terms had different but overlapping coverage.

• Suddenly, we had to deal with a large number of properties for which we had never asked.

We draw an important conclusion from this: LOD vocabularies are not means to describe our content models, i.e. they were not really useful for deriving domain models. We decided it would be better to interpret them as channels. That is, we model our information items in a Domain Ontology that is understandable by the domain experts. Interaction with them is essential to our approach and therefore understandability of our means towards these domain and communication experts. We then provide mappings (though our weaver, see Section 3.1.3) that export and/or import information to or from terms of various LOD vocabularies. Whenever we see a significant uptake of a vocabulary by a target group that we want to talk and disseminate to, we establish such a link. In the end, a term in a LOD vocabulary is treated similarly to a URI from our web pages. We export or import some of our content to or from it. For us, LOD vocabularies are means to disseminate and share information and not means to model information. Ontologies are always on the brink of being a very specific and well-defined domain model derived from certain first principles, being very useful for a specific purpose in contrast to broadly used and consensually developed models used for sharing information between different viewpoints. Consequently, we live in a world of multiple ontologies. “We no longer talk about a single ontology, but rather about a network of ontologies. Links must be defined between these ontologies and this network must allow overlapping ontologies with conflicting – and even contradictory – conceptualizations.” [7] We achieve this by weaving our models with LOD vocabularies when we see a gain in broadening our range of communication through them.24

3.1.2 Channel Model “A growing number of retailers are becoming increasingly multichannel as more

of their sales are coming through their web divisions than ever before.” [2]

“In telecommunications and computer networking, a communication channel, or channel, refers either to a physical transmission medium such as a wire or to a logical connection over a multiplexed medium such as a radio channel.”25 In on-line communication, we take a broad definition of a channel. A channel is a means of exchanging information in the on-line space. There is a close relationship between URIs and channels as each URI can be used as a channel to spread or access information. However, not each channel directly refers to an URI. For example, Facebook provides around forty different methods of spreading information not

                                                                                                               23 We had used a mélange of Dublin Core, FOAF, schema.org, and GoodRelations. 24 “The way the Semantic Web works, and this is what makes it very different from everything else, is that you

use a mixture of global ontologies like foaf:Person and dc:title and a number of other ontologies which are relevant, and then add on some more to make up what you need. If this sounds like a mess …” Tim Berners-Lee, email communication, Mon, 20 Feb 2012. Actually, we move the handling of this “mess” in the mapping of various vocabularies and free the user working at the information model level from it.

25 http://en.wikipedia.org/wiki/Communication_channel

  26  

distinguished by a URI. Additionally, individual information items spread through Facebook are not distinguished by URIs. In general, a channel can be interpreted as a “place” where one can find or leave information, whether it is unanimously referred by a URI or addressed through a service. However, even this is not broad enough. As described previously, a channel can also be the URI of a vocabulary (or the formalisms such as RDFa or microformats) that are used to publish the information. Through use of this URI, only humans or software agents that “speak” this dialect are able to access this information. Here, the communication channel cannot be interpreted as a place, but rather as a way to express or refer to the information. In the following, we want to distinguish channels by the communication mode they support. Communication is based on the broadcasting of information. Therefore, we define the first category of our channel classification system as channels used for broadcasting. Here we make a distinction between the publication of mostly static information and dynamic contents that express the timeliness of an information item. One way of spreading information is to invite other people to use it. Therefore, sharing is another category we have identified. It reflects the insight that others are not passive consumers of our information but active prosumers that should be helped and supported in their information processing activities. Sharing is the first form of cooperation. Explicit collaboration through a shared information space is the next cooperation category we have identified. Collaboration between individuals leads to groups of people actively organizing their communication and cooperation. Social networking sites that support groups of people in their information needs are instances of this next category we have identified. Obviously, the boundaries between these categories are fluid and many channel providers try intentionally to establish services covering several of them. Still, it is often possible to identify a major category for them, often based on the major usage patterns of their users. An important approach to broaden the scope of a dissemination activity is to add machine-processable semantics to the information. With this approach, search and aggregation engines can provide a much better service in finding and retrieving this information. A means of adding machine-processable semantics to information is our final channel category.26 Broadcasting static information. Websites are an established means of providing (mostly) static information. Information that reflects the structure of the contents is provided through websites and they offer a smooth way for users to access this content. An important addition beyond the dissemination through an owned website is an entry on other sites such as Wikipedia, the world’s leading encyclopedia. Broadcasting dynamic information. With Web 2.0 technologies, dedicated means for publishing streams and interacting with information prosumers have been added. A first step in this direction is the inclusion of a News section in a website using blogging tools such as Wordpress. Good practices for a news section on a website are:

• Each news item has its own URL, so that they can be returned in search results, bookmarked, shared etc.;

• News should contain a pointer to a more detailed description about the information items they describe;

• each news item is archived; • each news item can be indexed by search engines; • each news item is typed (through use of the information model); • each news item is categorized (through use of a folksonomy); • each post can be directly shared, emailed, added to favorites, and liked; • news can be searched, sorted, and filtered; and • important news items stay at the top to highlight main announcements. Such news can be further spread through a news ticker such as RSS feeds and Twitter. An

RSS feed is used to broadcast news. Its purpose is to regularly remind the user of the existence of a particular activity and the fact that it is producing interesting results. Twitter is a widely used means of disseminating news, however, significantly limits the length of it.                                                                                                                26 [30] propose a slightly different categorization of channels, however, call them different types of information.

  27  

Finally, Email and Email lists are also well established means for news dissemination. Especially the latter are a proven means of broadcasting information and facilitating group discussions. Other ways of spreading news are through social networking sites, which will be discussed below. Chatting is another form of instantly communicating and disseminating information, and a blog could be used to inform partners and members of recent trends in the field of semantic technologies. Sharing. There are a large number of Web 2.0 websites that support the sharing of information items such as: bookmarks, images, slides, and videos, etc. Collaboration. A wiki is primarily a means for project internal collaboration. However, it also becomes a dissemination channel if external visitors have read access27. They may then follow the intensive internal interaction that can help to gain a better and more detailed understanding of externally published results and achievements.

Group communication. Facebook as a social networking site provides an additional community aspect, i.e., it forms a community that multi-directionally shares news, photos, opinions, and other important aspects. Notice that Facebook is actually not only one, but several channels. It offers more than 40 possibilities through which to disseminate information. These can also be tightly integrated into Web 1.0 pages, such as that of the New York Times. 28 Google+ may have the potential to become a major competitor of Facebook. Therefore, it should also be included in a social networking site strategy. LinkedIn and Xing are focused on professional use and perfectly fit the purpose of research organizations.

Semantic-based Dissemination. An important approach to broaden the scope of a dissemination activity is to add machine-processable semantics to the information. With this approach, search and aggregation engines can provide much better service in finding and retrieving this information. Semantic annotations injected in websites are used by search engines such as Google to provide a structured presentation of the contents of websites, such as that shown in Figure 4, which can be analyzed by the format and vocabulary used. “This data may be embedded within enhanced search engine results, exposed to users through browser extensions, aggregated across websites or used by scripts running within those HTML pages.”[8] There are various formats of adding machine-processable semantics to data. First, there are three competing means of including semantics directly in HTML/XML files: (1) RDFa adds a set of attribute-level extensions to XHTML enabling the embedding of RDF triples; (2) Microformats directly use meta tags of XHTML to embed semantic information in web documents; (3) Microdata use HTML5 elements to include semantic descriptions into web documents aiming to replace RDFa and Microformats.29 For the moment, we have three competing proposals that should be supported in parallel until one of them can take a dominant role on the web. RDFa integrates best with the W3C meta data stack built on top of RDF. However, this also seems to hamper the uptake of this technology by many webmasters that are not familiar with this technology stack. Therefore, Microformats were developed as a competing approach directly using some existing HTML tags to include meta data in HTML documents. Actually, they overload the class tag which causes problems for some parsers as it makes semantic information and styling markup hard to differentiate. Therefore, Microdata instead introduce new tag attributes to include semantic data into HTML. Figure 5 shows that the use of RDFa has increased rapidly, whereas the deployment of microformats in the same period has not advanced remarkably. Consequently, we are focusing on RDFa and Microdata in our dissemination approach.

                                                                                                               27 Write access cannot be provided due to spamming. 28 http://www.nytimes.com/ 29 See [8] for more details.

  28  

 Fig.  4.  Structured  content  presentation30  

Instead of including semantic annotations in XHTML documents, i.e., injecting machine-readable contents into content that is meant for direct human consumption, they can also be provided for direct machine consumption. A straight-forward way is to publish an RDF file containing the machine readable data. Instead of directly publishing an RDF file you can also provide a SPARQL endpoint allowing the querying RDF information. Instead of retrieving the entire RDF file, directed queries can be supported with this approach In addition to predefined formats and technical means, we need to reuse predefined LOD vocabularies to describe our data to enable semantic-based retrieval of information.31

Currently, we use Dublin Core, FOAF, GoodRelations, and schema.org. Notice that we use each term of a vocabulary as a potential dissemination channel. For example, for the PlanetData fact sheet we publish pieces of the information using the following vocabulary terms: schema:url, foaf:topic, dc:creator, dc:date, dc:subject, and dc:title.

 Fig.  5.  Microformats  and  RDFa  deployment  on  the  Web  (%  of  all  web  pages)32  

3.1.3 Weaver The central element of our approach is the separation of content and communication channels. This allows reuse of the same content for various dissemination means. Through this reuse, we want to achieve scalability of multi-channel communication. The explicit modeling of content independent from specific channels also adds a second element of reuse: Similar agents (i.e., organizations active in the same domain) can reuse significant parts of such an information model. Separating content from channels also requires the explicit alignment of both. This is achieved through a weaver. Formally, a weaver is a set of tuples of nine elements:

1. An information item: As discussed in Section 2, it defines an information category that should be disseminated through various channels.

                                                                                                               30 Taken from http://www.google.com/support/webmasters/bin/answer.py?answer=99170 31 More than a hundred of them are listed at http://labs.mondeca.com/dataset/lov/index.html. 32 http://tripletalk.wordpress.com/2011/01/25/rdfa-deployment-across-the-web/

  29  

2. An editor: The editor defines the agent that is responsible for providing the content of an information item.

3. An editor interaction protocol: This defines the interaction protocol governing how an editor collects the content.

Elements 1 to 3 are about the content. They define the actual categories, the agent responsible for them, and the process of interacting with this agent. Elements 4 to 9 are about the dissemination of these items.

4. An information type: We make a distinction between three types of content: an instance of a concept, a set of instances of a concept (i.e., an extensional definition of the concept), and a concept description (i.e., an intensional definition of a concept).

5. A processing rule: These rules govern how the content is processed to fit a channel. Often only a subset of the overall information item fits a certain channel.

6. A channel: The media that is used to disseminate the information. 7. Scheduling information: Information on how often and in which intervals the

dissemination will be performed which includes temporal constrains over multi-channel disseminations.

8. An executor: It determines which agent or process is performing the update of a channel. Such an agent can be a human or a software solution.

9. An executor interaction protocol: It governs the interaction protocol defining how an executer receives its content.

First, the information types distinguish whether one wants to disseminate a general description of the information item, an instance of the information item, or a set of all instances. For example, we want to find an overall description of scientific presentations (what is their general theme) and a set of all presentations at a defined place on the web. The former may be placed on the project website and the later may be placed on SlideShare as a means to share presentations. Finally, a single instance may be broadcast as news through the various news broadcasting channels. Now, take a single presentation as an example. The title, author, abstract, and event it was given may form the news. The title, author, and a short notion of the event may define a tweet, and the slides themselves may go to SlideShare. That is, the information item must be processed to fit the various dissemination channels. A channel is a URI or an API of an existing web service. Scheduling information defines temporal constraints for dissemination in a single channel and for dependencies between multi-channel dissemination. For example, a new presentation will be announced once. However, an event may be announced as soon at it is defined and a reminder may be sent out when certain deadlines (for submitting papers or for early registrations) are near. News may first be published on the website. Then, an excerpt of the news together with its URI will be published as a tweet. A weaver is basically a large collection of tables specifying what is disseminated by whom to where. Interaction protocols, rules, and constraints further guide this process. Such a manual is of extreme importance to manage the on-line communication process. Obviously, it determines the need to implement and mechanize essential aspects of it, improving its scalability. However, a major step is to structure the process towards a mechanizable routine.

3.1.4 A Spiral as Process Model The basic SMCR model of communication33 is unidirectional. A sender sends a message through a channel to a receiver. The direction of the communication and the different roles are fixed. Actual communication is more complex. Agents interact and communicate in parallel, permanently alternating their role in these acts of communication. Therefore, we have adopted the transactional model of communication and its underlying premise that individuals are simultaneously engaging in sending and receiving messages (cf. [9]).34

                                                                                                               33 http://en.wikipedia.org/wiki/Models_of_communication 34 Or in Web2.0 terms, users are prosumers, i.e., consumer and producer of information.

  30  

Consequently, our approach not only disseminates information, but also deals with the aggregation of feedback and impact by simply going through the dissemination chain in the opposite direction, collecting responses in the various channels and integrating them under the appropriate knowledge item. We not only talk, we also listen to responses. And we do not get these responses scattered over multiple channels. Instead, they are aggregated and presented at the level of the domain specific concept to which they refer. Finally, our approach can also aggregate information from channels without prior publication activity. Communication situations initiated by an external agent can be handled through standing queries over pre-defined channels and dynamic web screening. Therefore, a holistic methodology for supporting communication must support the following subtasks that basically form a circle or spiral (see Figure 6):

• Design of an information item • Dissemination of the information item using suitable channels and places • Observation of communication acts • Measure, analysis, and aggregation of the information published

These activities form a circle that we call the life cycle model of communication. Reactive communication starts with the observation task; active communication starts with the design phase. In any case, when started, one has entered an-in principle-infinite loop. Obviously, these tasks can and should be parallelized once initialized. [10] identify five levels in reputation management:

• Listening: learn about the topic and the community; • Talking: participate in discussions; • Energizing: let other key players talk for you; • Supporting: support the community through initiatives and platforms; and • Embracing: the on-line community is used to further developing the company value

in a crowd sourcing approach.35 Each of these levels requires a different instantiation of our process model. In general, both models reflect the bi-directional character of communication in extension of more simplistic SMCR models.

 Fig.  6.  The  Life  Cycle  of  Communication  

3.2 Role Model

Editors are assigned to information items, responsible for producing or collecting their information instances. In general, an editor can also be the executor, directly publishing the information. However, expertise in a certain information domain may not necessarily

                                                                                                               35 See also [31] for a similar and refined model.

  31  

correspond to technical expertise and even if it does, it may not be a very efficient way of distributing labor. Only if a fully automated and easy to use software solution is provided can this model make sense. Otherwise, a person with more technical skills often helps in disseminating the information. Again, an interaction protocol has to be defined for interacting with this person. Recursively, some of his tasks may be manual (importing contents into a content management system such as Drupal) and some can be fully mechanized (like producing a feed and a tweet automatically for certain information items introduced into Drupal). We identified five different roles involved in this process (see Figure 7):

• The communication manager that actively reads and writes information in the multi-channel space and manages the overall communication process.

• The quality manager that routinely checks the outcome of the process and the impact that is achieved through it.

• The editors that provide information that should be disseminated or that infer actives from information provided by others.

• The web manager (executor) is an expert in web technology who is able to publish information with current web technology including content management systems such as Drupal, email lists and Web 2.0 services such as Twitter, Blogs, RSS, and has the means to share information, cooperate, or organize communities through SNS sites.

• The repository manager (executor) is an expert in semantic web technology in terms of syntax, implementations via repositories, and various vocabularies used to publish this information. In a nutshell, the web manager manages the web of documents and the repository manager manages the web of data.

 Fig.  7.  Roles  

3.3 Technologies

Obviously there is an important need for methods and integrated tools that cover the multi-channel bi-directional aspects of value management and provide highly scalable and effective

  32  

solutions. We are developing a communication platform and methodology for providing this based on Semantic Technologies, Human Computation, and Social Media Analysis.

3.3.1 Semantic Technologies Semantic Technologies (cf. [11]) are a stream of research combining web technology, artificial intelligence, natural language processing, information extraction, database technology, and communication theory for empowering computers to provide better support for processing, combining, and reusing information represented as structured and unstructured data. We use content management tools such as Drupal 7 to include RDFa, microdata, and microformats in the web documents. The data will also be exported from Drupal into OWLIM36 to support direct RDF37 and SPARQL (see Figure 8).

 Fig.  8.  Technical  means  to  publish  Semantic  Data  

Either the editors, or alternatively, the dissemination manger enrich the content for on-line presentation by adding links and tags to the presented information. For this purpose, tools such as the following are used: • KIM38 offers the ability to create semantic links between your documents, data, domain

models, and linked data; find mentions of entities, relationships, and facts in texts; and search and navigate your information space in multiple ways.

• The OpenCalais Web Service39 creates metadata for the content you submit. Using natural language processing (NLP), machine learning and other methods, Calais analyzes your document and finds the entities within it. Additionally, Calais returns the facts and events hidden within your text. The metadata gives you the ability to build maps linking documents to people, companies, places, products, events, geographies, etc. You can use those maps to improve site navigation, provide contextual syndication, tag and organize your content, create structured folksonomies, filter and de-duplicate news feeds, or analyze content to see if it contains information you care about.

                                                                                                               36 “OWLIM is a family of semantic repositories, or RDF database management systems, with the following

characteristics: native RDF engines, implemented in Java and compliant with Sesame, robust support for the semantics of RDFS, OWL Horst and OWL 2 RL, best scalability, loading and query evaluation performance” http://www.ontotext.com/owlim

37 The RDF file hast to be generated by OWLIM to include inferred triples. 38 http://www.ontotext.com/kim 39 http://www.opencalais.com/

  33  

• Zemanta40 is a platform for accelerating on-line content production for any web user. It searches the web for the most relevant images, smart links, keywords and text, instantly serving these results to the user to enrich and inform their content.

3.3.2 Human Computation Human Computation (cf. [12]) is a research field that aims to integrate human intelligence and human cognition into the computer-managed, automated execution of tasks that are beyond the power of the state of the art of pure algorithmic approaches. Until artificial intelligence has successfully enabled computers with sufficient human-like intelligence, on-line communication will continue to require and involve human labor. Understanding text, generating useful output, and interacting properly cannot be achieved without having a human in the loop. However, as we also saw it will not scale if too many tasks are left to humans. Therefore, we need to develop an architecture that tries to maximize the amount of tasks that are mechanized and that provides smooth integration of humans for tasks that cannot be fully automized. Such architecture may also have the potential to be applied in other domains with similar characterizations. This can include crowd-sourcing41 initiatives where we develop methods to define incentives for large user communities to provide content needed for your communication strategy through a distributed community effort.42

3.3.3 Social Media Analysis Social Media (cf. [13]) is a term used mostly for web-based techniques of human-to-human communication that stresses the social, topical, and contextual relations between communicating individuals, allowing real-time interaction with a large, yet specific audience of partners. Social Network Analysis emerged from the area of sociology that studies specific social phenomena. More recently, it was carried out under the umbrella term of complex network analysis, a field that studies properties in large, complex graphs. It became increasingly more popular through the huge success of social networks such as Facebook, Twitter, Flickr and others (cf. [15]). Web mining [14] is the use of data mining techniques to automatically discover and extract information from web documents and services. Machine learning, a branch of artificial intelligence, is a scientific discipline concerned with the design and development of algorithms that allow computers to evolve behaviors based on empirical data, such as from sensor data or databases.

For reputation management, we require support for impact analysis, multi-triggered bi-directional communication, and standing queries over streams. We enumerate some tools for this: • Boardreader43 is a search engine for Forums and Boards. • The ComScore Media Metrix Suite44 project is a framework for web traffic measurement

that is able to provide traffic statistics along several properties (age, region of users, etc.). Therefore, Media Metrix uses a representative set of Internet users, a weighting algorithm, and enumeration surveys to statistically make a projection of the whole population of Internet users. Additionally, ComScore introduced other tools such as mobile Internet traffic measurement and advertisement impact analysis software.

• Facebook Insights45 provides Facebook Page owners and Facebook Platform developers with metrics, just as CircleCount46 does for Google+. Google+ Ripples47 graphically illustrate the sharing of posts in Google+.

                                                                                                               40 http://www.zemanta.com/ 41 http://en.wikipedia.org/wiki/Crowd_sourcing 42 For example, see the results of the Insemtives project at http://www.insemtives.eu/ or the social stock market

empire avenue http://empireavenue.com/about/ 43 http://www.boardreader.com/ 44http://www.comscore.com/Products_Services/Product_Index/Media_Metrix_Suite/Media_Metrix_Core_Reports 45 http://www.facebook.com/help/search/?q=insights 46 http://www.circlecount.com/ 47 http://www.google.com/support/plus/bin/answer.py?answer=1713320

  34  

• Google Alerts48 are email updates of the latest relevant Google results (web, news, etc.) based on your choice of query or topic.

• Google Analytics49 is a web analytics solution that provides rich insight into website traffic and marketing effectiveness. It provides analysis and optimization tools such as Urchin50, Website Optimizer, Webmaster Tool51, Insights for Search52, and further tools for improving advertisement and Search Based Keyword Tools included in your website.

• Google Trends53 analyzes a portion of Google web searches to compute the number of searches that have been done for the terms you enter, relative to the total number of searches done on Google over time.

• ifttt54 provides a multi-channel trigger and action engine that allows automatic reaction to events in the information space by observing a multitude of channels and executing guarded transaction rules acting on these channels.

• Klout55 measures influence and impact of communication acts on social media. • Open Status Search56 finds out what people on Facebook are talking about in public. • PeerIndexes57 identifies, ranks, and scores on-line “authorities”. • The Social Media Dashboard58 manages multiple social profiles, allows scheduling

messages and tweets, can track brand mentions, and analyzes social media traffic. • Social Mention59 is a social media search and analysis platform that aggregates user

generated content into a single stream of information. Social Mention monitors 100+ social media properties directly including: Twitter, Facebook, FriendFeed, YouTube, Digg, Google etc.

• Technorati60 helps bloggers succeed by collecting, highlighting, and distributing the global online conversation. Founded as a blog search engine and directory, Technorati indexes more than a million blogs. Technorati tracks not only the authority and influence of blogs, but also the most comprehensive and current index of who and what is most popular in the Blogosphere.

• Twazzup61 operates a real-time news platform. • TweetDeck62 provides multi-channel integration for reading and writing for Twitter,

Facebook, MySpace, LinkedIn, Foursquare and Google Buzz • Twibes63 and Twitalyzer64 provide means to analyze content on Twitter.

3.4 Use Cases

We developed and applied our approach in three major case studies: the European Semantic Web Conference Series (ESWC)65, the PlanetData project66 and the Semantic Technology Institute (STI) International research association67.

                                                                                                               48 http://www.google.com/alerts/ 49 http://www.google.com/analytics/ 50 Urchin helps website owners better understand their online marketing initiatives, website traffic characteristics,

and visitors browsing experience. 51 Google Webmaster Tools provides you with detailed reports about your pages' visibility on Google. 52 With Google Insights for Search, you can compare search volume patterns across specific regions, categories,

time frames and properties. 53 http://www.google.com/trends 54 http://ifttt.com/ 55 http://klout.com/home 56 http://www.openstatussearch.com/ 57 http://www.peerindex.com/ 58 http://hootsuite.com/ 59 http://www.socialmention.com/ 60 http://www.technorati.com/ 61 http://www.twazzup.com/ 62 http://www.tweetdeck.com/ 63 http://www.twibes.com/ 64 http://www.twitalyzer.com/ 65 http://eswc-conferences.org/

  35  

• The mission of the Extended Semantic Web Conference (ESWC) series is to bring together researchers and practitioners dealing with different aspects of semantics on the web. Founded in 2004, the ESWC builds on the success of the former European Semantic Web Conference series, but seeks to extend its focus by engaging with other communities within and outside ICT, in which semantics can play an important role.

• PlanetData is a semantic technology project funded by the European Commission. It aims to create a durable community made up of academic and industrial partners working on large-scale data management.

• STI International is a global network engaging in research, education, innovation and commercialization activities on semantic technologies working to facilitate their use and applicability within industries and society as a whole. STI International is organized as a collaborative association of interested scientific, commercial and governmental parties that share a common vision.

Around 80% of the information items of ESWC, PlanetData, and STI International are interchangeable due to some simple renaming (e.g., core and associate partner versus partner and member). This is excellent news and a hint for scalability especially given the fact that we talk about a research project and a research association. This could imply that an even higher degree of reuse could be achieved when applying our information model to tens of thousands of European research projects (and hundreds of thousands of research projects or millions of projects) on the one hand, and millions of associations on the other. This is actually the second major assumption of our approach.68 Reuse of the information model in a certain vertical area. The costs to build an information models are quickly paid back when applicable to several entities in a domain. In the end, this is the SAP business model applied to on-line communication. These models empower simple non-IT users to communicate at the level of their domain knowledge rather than at the symbol level of various channels and these models can be reused between different players in the same vertical. Based on our approach ESWC, PlanetData, and STI International are now managing their on-line appearance. In total, we have identified around five hundred different semantic and non-semantic channels in these case studies that are used to disseminate elements of the information model. Obviously, such a bandwidth requires a structured and mechanized approach. Based on our approach, around 300 concepts and properties, 500 channels, i.e., more than 100,000 potential content-to-channel mappings are run efficiently by a very small dissemination team.

3.5 Our proposed solution in a nutshell

The core idea is to introduce a domain-specific, channel-independent model that explicitly separates content from channel. The next step is to once again intelligently interweave the content with the channels. Our approach of modeling communication, communication channels and target groups inherently bears the advantage of uniformly accessing the provided data and thereby allowing the processing of data that is beyond state of the art. For example, yield management could be realized utilizing reputation and usage values collected from different channels. Furthermore, the abstraction layer allows multi-channel communication. Human computation could increase the process where automated algorithms lack efficiency, for example the translation of communicated content into other languages. Combining these different areas of technology provides a long-term roadmap for research, engineering, and commercial exploitation.

                                                                                                                                                                                                                                                                                                                             66 http://www.planet-data.eu/ 67 http://www.sti2.org/ 68 The first one is that is that it will pay back to model the information independent from the multitude of

dissemination channels, ensuring reuse over them.

  36  

4 Related and Future Work

4.1 Related Work

Many aspects of our work clearly relate to different fields that have been explored before. Generally, we see two specifically related areas: Ontology-based content management systems (CMSs) for websites and Semantic matchmaking of senders and receivers of content. Both areas will be briefly described and compared. The field of semantics-based or enhanced CMSs has already been quite thoroughly explored. One of the earlier approaches to ontology-based website management is the OntoWebber system described in [16]. The proposed three-way approach of “explicit modeling of different aspects of websites", “the use of ontologies as foundation for Web portal design", and “semi-structured data technology for data integration and website modeling" presents an early but comprehensive approach to semantifying CMSs. OntoWebber introduces an integration layer which adapts to different data sources. This is related to our weaver concept introduced in Section 4, but, in contrast, the weaver adapts to different channels rather than to different information sources. A year later, in [17], Sheth et al. introduce the SCORE system, which defines four key features: semantic organization and use of metadata, semantic normalization, semantic search, and semantic association. Although written in the early days of the Semantic Web, the paper covers topics such as metadata extraction from unstructured text and automatic classification that may also become relevant to our approach. [18] introduce “The Rhizomer Semantic Content Management System” which integrates services with metadata browsing, editing, and uploading, continuing their earlier work on the Knowledge Web portal. [19] proposes a Linked Data extension for Drupal that enables content annotation with RDFa and provides a SPARQL endpoint. The British national broadcaster BBC started to integrate semantic technologies (i.e. Linked Data) in 2009 in order to integrate various data and content sources distributed throughout the enterprise [20]. As a result, as reported in [21], BBC's World Cup 2010 site69 is based on semantic repositories that enable the publishing of metadata about content rather than publishing the content itself. While the data input is fixed, different schemas for the output are defined. However, as only one channel for output is considered, the mapping performed is quite straight-forward. In contrast, our system accounts for different information needs of various and heterogeneous channels and therefore enables the distribution of content through different portals. Finally, the European project Interactive Knowledge Stack (IKS)70 focuses on porting semantic technologies to CMS software solutions. In a nutshell, all these approaches aim either to help the user publish semantic data or to use semantic methods to support the content management process for maintaining websites. We are taking these approaches and generalizing them to support the overall management of content dissemination in a multi-channel and bi-directional communication setting. Further, we augment the technical approach with a methodology and the approach of using vertical domain models, which are shared and reused in a vertical area instead of being used for a single application only. Semi-automatic matchmaking is a well-studied field in Artificial Intelligence and related areas. Obviously we can only select a small sample of approaches in this area, which focus on matchmaking in regard to content. [22] present a selective information dissemination system that is based on semantic relations. In their paper, the terms in user profiles and terms in documents are matched through semantic relations that are defined using a thesaurus. Similarly, the approach taken by [23] introduces selective dissemination of information for digital libraries based on matching information items to user profiles. Obviously, user profiles correspond to our channels, however, we instead manually model their relationship with contents. The system introduced in [24] uses RDF, OWL, and RSS to introduce an efficient publish/subscribe mechanism that includes an event matching algorithm based on graph                                                                                                                69 http://www.bbc.co.uk/worldcup 70 http://iks-project.eu/

  37  

matching. Our approach, in contrast, matches information items to channels rather than events to users. Also, instead of graph matching, we use predefined weavers for channel selection. While [23] uses fuzzy linguistic modeling and NLP techniques for semiautomatic thesaurus generation and performs a matching based on statistical analysis, we use semantics to manually define the connections between information items and the channels. Since we aim for high precision and professionalism in on-line communication, we see little use for statistical based semantic methods (natural language understanding, information extraction, etc.). We want to allow the user to abstract from the channel level to the content level, but we see the need for human involvement in defining the content-channel mapping and at the content level. However, as we expand towards a full-fledged value management approach that monitors the entire web space for important statements, such methods will be needed. Fortunately, a large number of such web analytical toolkits already exist, [4] lists a large number of them that cover parts of these tasks. However, there is an important need for methods and integrated tools that cover the multi-channel bi-directional aspects of value management and provide highly scalable and effective solutions. Obviously, the goal to develop a Common Value Management Framework (CVMF) based on combining these different areas of technology provides a long term roadmap for research, engineering, and commercial exploitation.

4.2 Future Work

Introducing a semantic layer on top of communication channels is required to enable a framework that allows common value management. However, this combination of research fields opens a broad variety of new challenges yet to be solved.

• Modeling and interweaving feedback. Feedback is an important part of all effective communication. Without feedback, the sender - the one who intends to convey information - has no means of validating whether or not the recipient received or understood the message. It is also often preferable to have a full-fledged two-way conversation instead of simple one-way broadcasting. The Web 2.0 revolution made it ridiculously easy for everybody to use the Internet as a two-way conversation platform where they can provide feedback as well as react to what was said. Therefore, it will be necessary to model feedback and interweave it with content items that we previously published.

• Modeling target groups. Companies that pursue common value management usually have a very restricted target group of people they wish to address. So far in our channel model, we do not distinguish between different target groups in different channels. However, different target groups reside in different communication platforms, even though there is some overlap. For example, you will find more young and hip people on Facebook and more professional users on Xing or LinkedIn, but there are quite a few users that have a profile on both platforms. Nonetheless, they expect a different way of being engaged in different platforms.

• Adapting content. Adapting content is a two-part problem - converting an information item into different formats or automatable transformations such as extracting images, videos or extracting and shortening web links from pieces of content, and transforming multimedia content into a different format. Both of these problems can be commonly solved, however, adapting content in a way that requires creativity and human intelligence is still a challenging problem that reaches the borders of computability. Examples of such adaptations are shortening or translating an essay, or rewriting a text in a way that matches the target group it addresses.

• Crowd sourcing. Crowd sourcing is an online approach of outsourcing tasks that computers cannot solve, to humans. There are different incentives for people to work on crowd sourcing problems. The most common one is small amounts of money (e.g. Amazon Mechanical Turk71, Clickworker72 or Zhubajie73) but other sources of

                                                                                                               71https://www.mturk.com/mturk/welcome

  38  

motivation are common such as competition (e.g. TopCoder74) or language learning (e.g. DuoLinguo 75). Crowd sourcing will be used to smartly enhance sentiment analysis, natural language processing, and translating algorithms with human intelligence where needed until artificial intelligence has reached a more satisfactory level. CrowdDB [25] is a successful example of combing the two computation paradigms.

• Crowdturfing and trust. Crowd sourcing solutions are very capable of dealing with hard to compute problems. However, this also allows for harming platforms that only defend themselves against automated attacks. Malicious activities, such as shaping opinions of a large number of people via social media platforms, and the use of crowd sourcing platforms are becoming more and more popular (cf. [26]). That is why some mechanisms of trust have to be introduced, which is also tightly connected to reputation management. A non-trustworthy source may communicate anything they want, the effect will be very little and its value drastically decreased.

• Quality management. An important part of targeted communication is assessing and improving the quality of conveyed content. Whereas trust, reputation, and brand management are influenced by how information is perceived, quality assurance is an inbound process. The business processes for quality management and what they actually mean, have yet to be defined for common value management. The bigger the campaign is, the more visible the effect of proper quality management.

• Quantification of social values. The success of online social networks leads to an enormous amount of data that may be analyzed in order to find out about social structures and relationships. A mature way of quantifying values like popularity, authority, influence or reach has yet to be found. Various tools such as PeerIndex76 or Klout 77 already attempt to measure it, but consider only very small parts of the whole social media landscape. Also, there are other attempts to measure various factors in social media using different techniques or different forms of proximity to forecast activity (cf. [27]).

• Quantification of brand and reputation. Similar to the quantification of social values, brand and reputation have to get a countable unit as well. Likewise, it will be very challenging to find fitting metrics, since already existing measures such as brand equity are considered meaningful by a small share of marketing professionals (cf. [28]). The combination with social media and the possibilities of sentiment analysis allows more suitable metrics to be introduced next to the existing ones.

• Enrichment of yield management. Yield management is based on statistical analysis on different parameters, such as pricing, capacity, and demand (cf. [29]). Already established calculation models can be extended by channel based reputation, brand value and other yet to be introduced criteria. For example, in a communication channel where a product’s brand (or the product itself) has little reputation and is badly represented, the price of the delivered service/product could be less than in other channels where it is well known. On the other hand, imagine a pricing model that not only considers capacity, but also takes social relationships into accounts, e.g. whether or not a popular person or a friend of yours purchased the offer.

It is evident that the long road of our journey still lies ahead.

5 Conclusions

The following core features characterize our approach:

                                                                                                                                                                                                                                                                                                                             72http://clickworker.com/ 73http://www.crowdsourcing.org/site/zhubajie/wwwzhubajiecom/2118 74http://www.topcoder.com/ 75http://duolingo.com/ 76http://www.peerindex.com/ 77http://klout.com/home

  39  

• We use ontologies to model content in order to have a representation layer independent from the communication channel. We want to achieve reuse of content over channels allowing small organizations to deal with an increasing number of communication channels and exploit their potential. The alignment of content and channel is achieved through a weaver that aligns ontological items with channels.

• These ontologies are not case-specific, but model a certain vertical domain such as research projects, associations, accommodations, restaurants, bars, touristic events and services, etc. Therefore, these ontologies and their channel alignments can be reused on a larger scale, providing a quick return of the investment necessary to build and maintain them.

• Our approach is bi-directional, i.e., in the same way that we disseminate through concepts we use these concept to aggregate feedback and impact found in various channels.

• We support in an integrated fashion, the dissemination via traditional web channels, Web 2.0, and semantic based channels, using various formats and vocabularies.

Based on our approach, ESWC, PlanetData, and STI International are now managing their on-line appearance. Around 300 concepts and properties, 500 channels, i.e., around 150,000 potential content-to-channel mappings are run by a very small dissemination team. Currently, we are performing additional case studies. First, we use our approach in the dissemination of other research projects and associations. Second, we are entering more commercial areas such as eTourim, where millions of hotels are desperately waiting for a scalable dissemination strategy, given the fact that soon, around 50% of all room bookings will be done on-line.  

Acknowledgements. We would like to thank Anja Bunnefeld, Johannes Breitfuss, Carmen Brenner, Alice Carpenter, Anna Fensel, Michael Fried, Mark Greaves, Marko Grobelnik, Martin Hepp, Lyndon Nixon, Ina O' Murchu, Alexander Oberhauser, Simeona Pellkvist, Elena Simperl, Corneliu Valentin Stanciu, Ioan Toma, and Alexander Wahler for contributions to early drafts of this paper.

6 References

[1] D. Amersdorffer, F. Bauhuber, R. Egger, and J. Oellrich (eds.): Social Web im Tourismus, Springer,

2010. [2] S. Mulpuru, H. H. Harteveldt, and D. Roberge: Five Retail eCommerce Trends To Watch In 2011,

Forrester Research Report, January 31, 2011. [3] THE BASICS OF REVENUE MANAGEMENT, Integrated Decisions and Systems, Inc., 2005.

http://www.adhp.org/pdf/1-theBasicsofRM.pdf [4] H. Kasper, M. Dausinger, H. Kett, and T. Renner: Marktstudie Social Media Monitoring Tools,

Frauenhofer Verlag, 2010. [5] S. Staab and R. Studer (eds.), Handbook on Ontologies, Springer Verlag, Heidelberg. 2nd edition, 2009. [6] T. Bauereiß, B. Leiter, and D. Fensel: Effective and Efficient On-lin Communication, STI TECHNICAL

REPORT 2011-12-01, 2011. http://www.sti-innsbruck.at/TR/OnlineCommunication [7] D. Fensel: Ontologies: Dynamic Networks of Formally Represented Meaning. 2001. http://sw-

portal.deri.at/papers/publications/network.pdf [8] J. Tennison: HTML Data Guide, W3C Editor’s Draft 02 March 2012. [9] D. C. Barlund: A transactional mode of communication. In C. D. Mortensen (eds.), Communication

Theory (2nd ed.), New Jersey Transaction, 2008. [10] C. Li and J. Bernhoff: Groundswell: winning in a world transformed by social technologies, Forester

research, Harvard Business School Publishing, 2008. [11] Domingue, J., Fensel, D., & Hendler, J. A. (2011). Handbook of Semantic Web Technologies (Vol. 1).

Springer. [12] L. Ahn v. (2005). Human Computation. Ph.D. dissertation, CMU-CS-05-193. [13] A. Kaplan, and M. Haenlein. (2010). Users of the world, unite! The challenges and opportunities of

Social Media. Business horizons , 53 (1), 59-68. [14] R. Kosala and H. Blockeel. (2000). Web mining research: A survey. ACM Sigkdd Explorations

Newsletter , 2 (1), 1-15. [15] F. Bonchi, C. Castillo, A. Gionis, and A. Jaimes. (2011). Social Network Analysis and Mining for

Business Applications. ACM TIST , 2 (3), 22. [16] Y. Jin, S. Decker, and G. Wiederhold. Ontowebber: Modeldriven ontology-based website management.

In Proceedings of the Semantic Web Working Symposium (SWWS), Stanford University, 2001.

  40  

[17] A. Sheth, C. Bertram, D. Avant, B. Hammond, K. Kochut, and Y. Warke: Managing semantic content for the web, IEEE Internet Computing, 6:80-87, July 2002.

[18] R. Garcia, J. M. Gimeno, F. Perdrix, R. Gil, and M. Oliva: The Rhizomer Semantic Content Management System. In Proceedings of the 1st World Summit on The Knowledge Society: Emerging Technologies and Information Systems for the Knowledge Society, WSKS’08, pages 385-394, Springer, 2008.

[19] S. Corlosquet, R. Delbru, T. Clark, A. Polleres, and S. Decker: Produce and consume linked data with Drupal, Springer Constraints Journal, 1380:763-778, 2009.

[20] G. Kobilarov, T. Scott, Y. Raimond, S. Oliver, C. Sizemore, M. Smethurst, C. Bizer, and R. Lee: Media meets semantic web: how the BBC uses dbpedia and linked data to make connections. In Proceedings of European Semantic Web Conference (ESWC 2009), LNCS 5554, pp. 723-737, 2009.

[21] B. Bishop, A. Kiryakov, D. Ognyanoff, I. Peikov, Z. Tashev, and R. Velkov: OWLIM: A family of scalable semantic repositories, Technical report, 2010.

[22] I.-E. Katzagiannaki and D. Plexousakis: Information dissemination based on semantic relations. In CAiSE Short Paper Proceedings'03, 2003.

[23] J. M. Morales-del-Castillo, R. Pedraza-Jimenez, A. A. Ruiz, E. Peis, and E. Herrera-Viedma: A semantic Model of Selective Dissemination of Information for Digital Libraries, Information Technology and Libraries, 28(1):21-31, 2009.

[24] J. Ma, G. Xu, J. L. Wang, and T. Huang: A semantic publish/subscribe system for selective dissemination of the RSS documents. In Proceedings of the Fifth International Conference on Grid and Cooperative Computing (GCC'06), pp. 432-439, 2006.

[25] M. J. Franklin, D. Kossmann, T. Kraska, S. Ramesh, and R. Xin: CrowdDB: Answering queries with crowdsourcing, SIGMOD '11 Proceedings of the 2011 international conference on Management of data, 2011.

[26] G. Wang, C. Wilson, X. Zhao, Y. Zhu, M. Mohanlal, H. Zheng and B. Y. Zhao: Serf and Turf: Crowdturfing for Fun and Profit CoRR, Vol. abs/1111.5654, 2011. http://arxiv.org/abs/1111.5654

[27] K. Lerman, S. Intagorn, J.-H. Kang and R. Ghosh: Using Proximity to Predict Activity in Social Networks. CoRR, Vol. abs/1112.2755. http://arxiv.org/abs/1112.2755

[28] P. W. Farris, N. T. Bendle, P. E. Pfeifer, and D. J. Reibstein: Marketing Metrics: The Definitive Guide to Measuring Marketing Performance (2nd Ed.), Pearson Education, 2010.

[29] S. E: Kimes: A strategic approach to yield management. In A. Ingold, U. McMahon-Battie, and I. Yeoman (eds.), Yield Management: Strategies for the Service Industry, International Thomson Business Press, 2000.

[30] J. Finzen, H. Kasper, and M. Kintz: Innovation Mining, Frauenhofer Verlag, 2010. [31] T. Helbing and M. Konitzer: Das Ohr am Puls des Internets. In [Amersdorffer et al. (eds.), 2010].

 

 

 

  41  

 

Diversity through Social Media

Tadej Štajner1 and Inna Novalija1

1Artificial Intelligence Laboratory, Jožef Stefan Institute, Ljubljana, Slovenia {tadej.stajner, inna.koval}@ijs.si

Abstract. This paper describes the approach for interactive diversity analysis based on social media data. The presented approach is applicable for a number of common value management activities, such as reputation management, competition analysis, market analysis, sentiment analysis and similar tasks, related to measuring community activity. The design of the developed diversity analysis component allows achieving the quickest possible turnaround time required to train a multi-class model for a particular task. The evaluation is performed in the telecommunication domain on a dataset of social media microposts.

Keywords: Natural language processing, Diversity analysis, Sentiment analysis, Social networks.

1 Introduction

This paper describes the approach for diversity analysis based on social media data. Diversity analysis represents a collection of related tasks that segments posts and users into meaningful groups, based on differing viewpoints that depend on the topic, the users’ location and language, providing sentiment and opinion modeling given these groups [1]. Diversity analysis can be applied for a number of brand management activities, such as reputation management, competition analysis, market analysis, and sentiment analysis. These tasks have a goal of establishing, monitoring and improving the status of subject (for instance, a product or a particular organization) in the community.

We motivate our work with the case study on customer relationship management through social media, especially various forms of characterization of the retrieved relevant posts. Since social media is consistently becoming an important communication channel for brand and customer relationship management, new technical issues are coming up when we try to apply previously known techniques to communication through social media. Imagine that our user, working in customer relationship management or marketing, is trying to measure the impact and opinions on a new campaign. In this scenario, we emphasize the use case of making sense of the social media responses by having the user retrieve and segment tweets relevant to the campaign in question. More specifically, the scenario discussed in this paper is the following: out of a stream of social media posts, assign classes to the posts that are relevant to the brand management task at hand, where the task can be unknown in advance. Furthermore, by classifying these posts we are annotating them with concepts from a light-weight ontology [6], a use case that has already successfully demonstrated using active learning techniques for a related task, namely ontology construction.

In this research we present an approach that can be used to train one of several ad-hoc model-based queries that can appear in such an environment: language, topic or sentiment classification. Since brand management requires constant maintenance, it’s difficult to foresee in advance what exact problem we are tackling. Therefore, most such problems need to be solved ad-hoc. In order for that to be practical, they also need to be trained quickly. The key contribution of this research is the demonstrated ability to interactively train such a model in order to answer these types of questions, providing quick analyst response. For the purpose of

  42  

this paper, we frame the problems as multi-class classification tasks, where we also allow for unlabeled examples. There are several relevant problems that fit this framework. While some of these problems can be defined in advance, such as coarse topic classification, language detection or general sentiment detection, some can be predictable, for instance the launch of a new product, a new competitive product or a community-driven action. While we can initially approach this problem using information retrieval techniques, formulating a query may be difficult and ambiguous, which is why we can also consider this problem in terms of multi-class classification.

We focus on the telecommunication domain, the tweets about different competitors – telecommunication companies are labeled with different topics (Vodafone, O2, Simobil), and we wish to evaluate the overall sentiment in social media with respect to these competitors. The sentiment classification contributes to the activities of reputation management and sentiment analysis. However, due to topic-sensitivity of sentiment analysis, the there is a need to produce specialized sentiment models.

The design of the developed diversity analysis software components allows achieving the quickest possible turnaround time required to train a multi-class model for a particular task. Some tasks originate from a particular marketing or public relations project on a particular new offering, whose public opinion needs to be monitored, which call for ad-hoc definition of tasks. We also need the changes are immediately apparent in the result set.

This paper also describes this approach in detail, the respective design decisions and demonstrates an example software component. The approach and software evaluation is performed in the telecommunication domain on a dataset of Twitter posts.

2 Related Work

There are a number of approaches discussing diversity analysis. In their work Hasan et al. [1] explained the baselines of the knowledge diversity model and presented the glossary of diversity relevant terms. They present three motivating business scenarios for formalization of the knowledge diversity model – mining diversity in Wikipedia, diversified news and customer relationship management. The diversity in opinions has been discussed by Kim and Hovy [1] in their paper on opinion sentiment. They presented a system that, given a topic, automatically identifies people with opinions (and opinion sentiments) on that topic.

Turney [2] developed an unsupervised algorithm for classifying reviews as recommended or not recommended based on the average semantic orientation of the phrases in the reviews.

At the same time, social media have been widely used for opinion determination. For instance, Bizău et al. [4] presented an approach to creating domain dependent opinion vocabulary based on Twitter comment, acknowledging the diversity that heavily affects that particular problem.

In our research, social media data is used as a corpus for common brand management activities - reputation management, competition analysis, market analysis and sentiment analysis. We are not only combining different brand management activities in one tool – we are also minimizing the turnaround time required to train a multi-class model for a particular task, while letting the user to track the quality of the obtained model in order to estimate whether more training is necessary.

  43  

3 Introduction

To accommodate the functional requirements for brand positioning in social media, we have developed an approach that enables interactive ad-hoc training of tasks such as topic, language and sentiment detection as semi-supervised multi-class classification tasks. The developed software component is intended to be the analytics back-end for a social media analytics dashboard. The situation we are dealing with is providing social media analytics to answer ad-hoc model-based queries on social media streams. The main problem is that the model may not be known in advance, so it needs to be trained quickly.

For instance, for a telecommunication domain, the model should answer the following queries:

What kinds of sentiment do people express on O2’s iPhone offer?

Which operator has the most social media talk about Android devices?

In order to train a specified model, we use a machine learning technique called active learning, a technique that intelligently picks unlabelled examples which need to be labeled in order to minimize training time.

At the same time, we encounter a number of the technical design constraints:

• We have multiple classes to classify in; • We can’t realistically expect the user to label all examples; • We should make use of feature distribution using all available (even unlabelled) data; • The model changes should propagate in real-time with respect to labeling.

The design of the software component was driven to achieve the quickest possible turnaround time required to train a multi-class model for a particular task. While some tasks are pre-defined, many of them come from a particular marketing or public relations project on a particular new offering. Therefore, we allow the ad-hoc definition of tasks. We also focus on providing real-time training, so that new labels are quickly incorporated in the model.  

3.1 Internal domain model

First, we describe the concepts of the diversity analysis software service. In order to accommodate several different tasks under the same umbrella (i.e. language, topic and sentiment detection), we operate with the following concepts: • A task is a concept that describes the goal that we wish to achieve. Each task has multiple

target classes. We allow examples also to not belong to any class. For example, language detection has each language as a target class, topic classification has topics and sentiment classification has positive and negative, along with neutral.

• A model is a particular approach configured to execute a task. A single task may have multiple models. Each model can have a single model algorithm, which defines the concrete approach for the classification (for instance SVM or Naïve Bayes).

• An item is a single data point, consisting of a set of features. This may be a single tweet or a survey record. Each item may have an associated source, language, author and geographic location.

• A label indicates that a particular item is associated with a particular target class in a particular task. The labels are used for training and evaluating the models.

  44  

For example, imagine that we have a sentiment detection task, which is a three-way classification problem (positive, negative, and neutral). Then, we declare a model using Naïve Bayes. In order to train the model, we use the search functionality to retrieve examples that we consider to be relevant for the model. Given a query “blackberry”, we obtain the following tweet (among others):  

I LOVE THE NEW BLACKBERRY TV COMMERCIAL. COMING AFTER BLACKBERRY BOYS CAMPAIGN BY VODAFONE, IMAGE MAKE OVER GUARANTEED FOR BLACKBERRY

When the model is still fresh and untrained, none of the examples are labeled. Since we wish to train the model, we assign a positive label to this example. That label gets added to the task and is subsequently used to retrain the model.

3.2 External domain model

While the system is general enough to map to any multi-concept annotation task, it is primarily meant to be applied to fit into the Knowledge Diversity Ontology (KDO) [11], which is designed to represent the diversity of opinions and mentions on the web, a domain which is relevant for describing situations in customer relationship and brand management. The classes in task examples that we are demonstrating can all be mapped to instances of KDO concepts.

Specifically for the tasks that we are demonstrating, we can map them to the following RDF statements: the language classification asserts an xml:lang property of the post. Topic classification is a task of annotating posts by asserting that they have a sioc:topic relation with a certain topic. In more specific classifications where we are considering detecting concrete entities, we recommend using the kdo:mentions predicate. In the sentiment detection case, we are annotating the post with the kdo:hasSentiment predicate, asserting that the post has a sentiment (kdo:Sentiment) of a given polarity (kdo:hasPolarity, kdo:Polarity), which can be one of kdo:positivePolarity, kdo:negativePolarity or kdo:neutralPolarity.

3.3 Description of approach

As stated above, in order to achieve good turn-around time, we employ active learning, which allows data instances to be labeled for training by an oracle (usually a human annotator). It is now a well-established finding that such methods can achieve higher accuracy with fewer training examples than passive learning [5]. However, these techniques often suffer from the cold-start problem: when no or few examples are labeled, it is difficult to get quality queries from the active learning mechanism. Therefore, we supply an alternative method for retrieving examples: besides offering to label uncertain examples, we also allow the user to supply a textual query that retrieves relevant items using a search index, a method which was also used in the OntoGen system in order to quickly define new ontology concepts [6]. If a model is available, we also use it to label the results of the search query.

The key hypothesis of active learning is that if the algorithm is allowed to choose the data from which it learns, it can perform better with less learning. Then, the chosen data points are submitted to the oracle as queries, which the oracle then labels. Active learning algorithms therefore need to have the ability to rank the unlabelled examples with regard to the expected utility they can bring to the model. In practice, there are several approaches to estimate that.

When using Support Vector Machines to train the classifier, one possible method to select the examples for labeling is to pick the examples which lie the closest to the hyperplane in the

  45  

feature space that separates the classes [7]. In a multi-class setting, we can use the sum of distances to the separating hyperplanes. Intuitively, the examples that are close to the hyperplane that separates classes tend to be the ones which are less certain to belong in one or the other class. Therefore, it makes more sense to label those data points than labeling them randomly.

When using Naïve Bayes to train classifier, we can employ the uncertainty sampling approach [8]. In this framework, the active learner recommends the instances about which it is the least certain on how to label, which is straightforward for probabilistic learning models, such as Naïve Bayes. For multi-class learning, this can be generalized to recommend the instance whose prediction is least confident over all classes.

We employ basic feature extraction: the word tokens and bigrams are extracted from the content, stripped of case and used in a bag-of-words model.

3.4 Architecture

 

 

 

 

 

 

 

 

 

 

Fig. 9. Architecture of the interactive diversity analysis service.

The architecture of the component is displayed in Figure 9. While the diagram also includes a graphical user interface built for the purpose of demonstrating the system, the primary method of interface is through the Interactive diversity analysis API. In this work, we demonstrate the functionality of the component with a HTML5 client that directly connects to the API and is in no way privileged than other potential consumers of the component.

The back-end is driven by an embedded relational database to store the items and metadata. The search indexing and feature construction part is driven by JSI’s Miner infrastructure to ensure handling of datasets bigger than available main memory. The indexing and retrieval back-end is the same that has also been used in other use cases, such as real-time news recommendation [9]. The models themselves are stored in-memory, since they need to be re-trained often. This does not pose significant challenges for scalability, since the required storage from the model is usually proportional to the size of the feature space and not item count. The system also utilized caching of generated feature vectors for items to enable efficient re-training of models when new labels arrive.

Documents, Metadata

Indexes, Models (JSI Miner)

Diversity analysis component (JSON services)

GUI (HTML5)

Other integrating

components

  46  

The main controller of the system is taking care of the maintenance of models. Since we require real-time data updates and immediate effect of labeling without blocking the user’s requests, we have implemented concurrent model training. For instance: when the system receives a new label, it first checks if the model is currently in the process of re-training. If it is, the label is put in a queue for the next pass. This way, adding a label is a non-blocking process with regard to model training. When the model finishes training, the new re-trained model immediately replaces the old model, so that all subsequent classifications are executed with the new model. This ensures that also the search operation is non-blocking with regard to model training.

Usually, the training time is on the order of a second for several thousand examples – just long enough to be a perceptible delay in blocking mode. In non-blocking mode, this delay is barely noticeable, since adding a label and issuing a new search query with classification are two separate interactions, and the re-training usually finishes in the time before the user starts with search and classification.

3.5 User interface

Figure 10 displays the graphical user interface that demonstrates the functionality of the diversity analysis service, built in HTML5 and using jQuery for display logic. It has four main sections: the query configuration box, the result set, the uncertain examples and at the bottom, the configuration panel. The example shows the state after the user has issued the query ‘blackberry’ in the sentiment classification context, which caused all results to be labeled according to the model. The main use case is the following: after the user defines the task and the model, she then issues a query that is considered to be most informative for that particular task. After the result set is displayed, each result is equipped with a set of buttons that act as labeling triggers: pressing one of them acts as adding a label into the model. If the user repeats the query after labeling some examples, he may notice that the results may be re-grouped and re-ordered. This is the consequence of online updates of the model and can also be used to correct the algorithm by labeling misclassified examples.

However, there are several strategies by which the models are trained. While the first strategy requires issuing a query and labeling the results as described in the previous paragraph, the second one is by looking at the uncertain examples that are provided in on the right-hand side of the UI. These examples are provided by the active learning mechanism and they are sampled with proportion to their classification uncertainty given the model. In probabilistic models, such as multi-class Naïve Bayes which is used in this implementation, the uncertainty is equal to the total uncertainty over all classes.

However, due to large class imbalance, which is often an issue in brand management tasks that we encountered, the uncertainty sampling approach suggested examples that were often unrelated to the task at hand. Therefore, we devised another strategy to mitigate that. The third strategy is a hybrid of both: it samples the most uncertain examples that satisfy the search query.

The behavior of the user interface is therefore the following: if the user issues a query without a model, the center of the page displays search results without any classification. If a user is working in the context of a particular active learning model, try to classify the search results in real-time and also provide the uncertainty samples that match the query. Thirdly, if the user issues a blank query, the system just provides uncertain examples in the right-hand blue box without any restrictions on a query.    

  47  

 

Fig 10. HTML5 front-end that demonstrates the functionality of the diversity analysis service

4 Evaluation

The evaluation of the proposed approach and developed software is performed in the telecommunication domain on the dataset of Twitter posts. We are using a dataset of 14000 tweets gathered during one day.

We have created 3 scenarios, connected to common value management activities:

Scenario 1: Telecommunication company reputation management. For a particular telecommunication company (“Vodafone”) we have created a model, which displays the tweets related to company activities. Using active learning techniques, we have labeled a

Query  configuration  Exam

ples  retrieved  by  uncertainty  or  class-­‐margin  

sampling  

Query  result,  grouped  by  predicted  

label  

Manual  model  management  

Manual  task  management  

  48  

number of posts with positive, negative and neutral sentiment and then checked company reputation based on the created model.

The examples of positive and negative tweets about “Vodafone” are presented below:

 

 

 

Fig. 11. Examples of positive and negative tweets for Vodafone company

After labeling 50 posts, we have obtained the precision of 42 % and the recall of 50 %. Increasing labeling to 300 posts, we got the precision of 69 % and a recall of 63 %. In such way, active learning technique contributes to better results for our reputation management scenario.

Scenario 2: Competition analysis in the telecommunication domain. For a number of different telecommunication companies (“Vodafone”,”O2”) we have created a model, displaying the variety of opinions from Twitter users about the competitors.

Using active learning techniques, we have labeled twitter posts with relative posts for a particular competitor.

For a query “customer” we have observed how user opinions vary among the competitors. Fig. 12 provides the examples of different Twitter posts for competitors in order to spot weaknesses in their customer relationship management.

 

 

 

 

 

 

 

 

Positive  tweet:  

 

VODAFONE  INTRODUCES  CUSTOMERS  TO  THE  BLACKBERRY  TORCH:  IT'S  AN  EXCITING  DAY  FOR  VODAFONE  MALTA  A...  HTTP://BIT.LY/CIK0P6  #VODAFONE  Negative  tweet:    

 

VODAFONE:  YOUR  WIFE'S  PHONE  MAY  NOT  BE  COVERED  SIR.  ME:  WHY?  VODAFONE:  COZ  SHE  DROPPED  IT.  ME:  WHAT  DO  YOU  COVER?  VODAFONE:  NOT  DROPPING  IT.  

 

 

O2:  

 

O2,  BEEN  A  CUSTOMER  OF  THEIRS  FOR  EIGHT  YEARS  AND  YET  I  PAY  MORE  THAN  A  NEW  CUSTOMER.  TOSSERS  

 

JUST  HAD  THE  RUDEST  O2  CUSTOMER  SERVICE  MANAGER  EVER  ON  THE  PHONE.  THEY  NEED  TO  WORK  ON  THEIR  CUSTOMER  CARE  COZ  I  AM  SO  LEAVING  THEM.  

 

  49  

 Fig. 12. Examples of Twitter posts for different competitors

Scenario 3: Market analysis in the telecommunication domain. For a selected topic (“BlackBerry”), we have created a model, displaying user opinions about this topic in different languages (English, Spanish, and German). Using active learning techniques, we have labeled 20 posts as relative for the selected topic in a particular language. In such way, we were able to obtain user opinions in several markets – Spanish speaking market, English speaking market and German speaking market. Fig. 13 demonstrates the examples of Twitter posts for English and Spanish languages.

 

 

 

 

 

 

 

 

 

 

Fig. 13. Examples of Twitter posts for different languages

With having more posts in English, after labeling 20 Twitter posts, the precision for English language constituted 53%. The recall for English language was 76 %. While state-of-the-art language detection techniques achieve higher scores (with more training data), our purpose was to demonstrate that even low training numbers can achieve reasonable performance.

5 Conclusion

In this paper we described the approach for diversity analysis based on social media data. We have shown how diversity analysis can be applied for a number of common value management activities, such as reputation management, competition analysis, market analysis, sentiment analysis. The presented approach can be used to train a language, topic or sentiment classifier.

Vodafone:  

 

OF  FRUSTRATION:  DEALING  WITH  #VODAFONE  CUSTOMER  SERVICE.  I  HOPE  YOU  ARE  LISTENING,  #VODAFONE  -­‐  YOUR  CUSTOMER  SERVICE  SUCKS  

 

A  NICE  CUSTOMER  SERVICES  MAN,  HE'S  GONNA  SIT  THERE  ALL  DAY  TO  GET  ME  CONNECTED  TO  UPGRADE  SERVICES  N1  #VODAFONE  'S  CUSTOMER  SERVICE.  

 

English:  

 

LOVE  THE  NEW  BLACKBERRY  TV  COMMERCIAL.  COMING  AFTER  BLACKBERRY  BOYS  CAMPAIGN  BY  VODAFONE,  IMAGE  MAKE  OVER  GUARANTEED  FOR  BLACKBERRY.    

GET  FREE  BLACKBERRY  INTERNET  WITH  VODAFONE  PAY  &  GO  UNTIL  30  JUNE  2011  WE  HAVE  SIMS  AVALIBLE  SO  DROP  IN  FOR  FREE  BLACKBERRY  INTERNET  

Spanish:  

@SAJIID_GAGA  HACIENDO  BERRINCHE  POR  BLACKBERRY  MSN  VENTE  A  MOVISTAR  UNIDOS  HACEMOS  MAS  JAJA...  Y  EL  COMERCIAL?  LO  SIENTO  AMO  MI  TRABAJO  JAJA  

MOVISTAR  SOLO  TARDO  MINUTOS  PARA  ACTIVARME  EL  PLAN  BLACKBERRY..  DIDITEL  TARDO  3  DIAS  :/  

 

  50  

This paper also described an example client of the software component. The created software allows working not only with pre-defined tasks, but also with ad-hoc definitions of tasks. The approach and software evaluation is performed in the telecommunication domain on the dataset of Twitter posts.

There are still several requirement points that need to be addressed. Most notably, we need to support composition of models via search facets, for example supporting queries like “search only for English O2 tweets and obtain their sentiment”, as well as drill-down search. Another required feature is keyword suggestion, which can be achieved via query expansion and keyword extraction techniques. Besides that, we plan to also include support of structure extraction in opinions, a task appropriate for relational clustering and visualization methods.

When observing the model performance, we noticed that there is room for improvement in several directions for our future work. Firstly, the current implementation still does not take into account the potentially vast quantity of unlabeled data using semi-supervised algorithms, possibly further improving the convergence of the active learning workflow. Secondly, we can also improve performance by applying a transfer learning technique called multi-task learning [10], where one can train models for multiple related tasks simultaneously, having them improve each other.

Furthermore, there are several situations where we need to answer model-based queries which are special cases of existing queries. While many of these can be made via simple composition of model classifications and attribute-based criteria, there are some cases where model specialization is necessary. For example, we could use a general sentiment classifier in order to bootstrap another sentiment classifier for a very specific topic. For instance, training a sentiment classifier on the topic “pre-paid service for Movistar in Spanish” can be bootstrapped with the general Spanish sentiment model and specialized with additional annotations.

There are still open issues in the design of the classifier: for the sake of demonstration simplicity, the feature construction is currently the same for all of the tasks. Having a possibility to add new feature sets at runtime would definitely improve performance on several of the mentioned tasks.

Furthermore, since the algorithm is designed to work on streams of data, along with streams of correction inputs, it is also a potential scenario for application of online learning algorithms. Another interesting application of these techniques can also focus on providing global and aggregate results, not only samples that match queries.  

6 Acknowledgements

This work has been partially supported by the project RENDER under the grant number FP7257790 funded by the Seventh Framework Programme of the European Comission and the Slovenian Research Agency.

7 References

[1] Hasan, R., Siorpaes, K., Krummenacher, R., Flöck, F.: Towards a Knowledge Diversity Model, In: DiversiWeb-2011, Knowledge Diversity on the Web (2011).

[2] Kim, S-M., Hovy, E.: Determining the sentiment of opinions. In: Proceedings of COLING (2004). [3] Turney, P.D.: Thumbs up or thumbs down? Semantic orientation applied to unsupervised classification of

reviews. In: Proceedings of the 40th Annual Meeting on ACL (2002). [4] Bizău, A., Rusu, D., Mladenić, D.: Expressing Opinion Diversity. In: DiversiWeb-2011, Knowledge

Diversity on the Web (2011). [5] Settles, B.: Active Learning Literature Survey. Computer Sciences Technical Report 1648, University of

Wisconsin-Madison.

  51  

[6] Fortuna, B., Grobelnik, M., Mladenić, D.: Ontogen: Semi-automatic ontology editor. Human Interface and the Management of Information. Interacting in Information Environments, pp. 309-318, Springer (2007).

[7] Tong, S., Koller, D.: Support vector machine active learning with applications to text classification. The Journal of Machine Learning Research, vol 2., pp. 45-66, JMLR.org (2002).

[8] Lewis, D.D., Gale, W.A.: A sequential algorithm for training text classifiers. In: Proceedings of the 17th annual international ACM SIGIR conference on Research and development in information retrieval (1994).

[9] Fortuna, B. and Fortuna, C. and Mladenić, D.: Real-time news recommender system, Machine Learning and Knowledge Discovery in Databases, pp. 583-586, Springer, 2010.

[10] Pan, S.J. and Yang, Q.: A survey on transfer learning, Knowledge and Data Engineering, IEEE Transactions on, vol 22, no. 10, pp. 1345-1359, IEEE, 2010.

[11] A Thalhammer, I. Toma, R. Hasan, E. Simperl, and D. Vrandecic, How to Represent Knowledge Diversity, Poster at 10th International Semantic Web Conference, 2011.

 

 

 

 

  52  

An approach for evaluation of social media monitoring tools

Ioannis Stavrakantonakis1, Andreea-Elena Gagiu1, Harriet Kasper2, Ioan Toma1, and Andreas Thalhammer1  

 

1  Semantic Technology Institute (STI) - Innsbruck

ICT Technologiepark, Technikerstrasse 21a, 6020 Innsbruck, Austria {firstname.lastname}@sti2.at

2  Fraunhofer Institute for Industrial Engineering IAO

Nobelstr. 12, 70569 Stuttgart, Germany harriet.kasper@iao.fraunhofer.de

Abstract.  Social networks are important means for communication, engaging millions of users around the globe. For enterprises in particular, being present and aware of what is discussed on these communication channels about their products and services has become a must. Social media monitoring tools enable enterprises to have access to real customers’ opinions, complaints and questions at real time in a highly scalable way. As the number of social monitoring tools has rapidly increased in the last years, enterprises are faced with the difficult tasks of choosing the right tool for their needs. This paper proposes a structured evaluation framework comprising a set of evaluation criteria that can be used to analyze social monitoring tools from three perspectives: the concepts they implement, the technologies used and the user interface they provide. To exemplify the usefulness of our evaluation framework we analyze a set of social monitoring tools after briefly describing them.

Keywords:  social media monitoring, listening platforms, business intelligence

8 Introduction

Given the increasingly large number of consumers using social media, enterprises cannot ignore the power that is weaved within its networks. For instance, Power Reviews statistics on social commerce stats state that there are 500 million active Facebook users, 65 million tweets and over 3 billion Google searches each day [16]. People are using social media networks to express their needs and complaints, as well as opinions about proprietary products and services, and to compare them with solutions from other vendors. Due to the explosion of social media sites, marketers have an exponentially larger audience and the ability to instantly communicate with consumers [12]. In this respect, Forrester Research forecasts that, in terms of spending, social media marketing will reach an annual growth rate of 34%, outmatching all other forms of online marketing [13].

Enterprises utilize a wide range of traditional and nontraditional methods to listen to customers; however, in recent years, survey researchers are facing difficulties in collecting data through the traditional methods due to the decrease in landline telephone coverage and willingness of respondents to participate [14]. Moreover, the attractiveness of using free online sources of information is further sustained by the relative costliness and time-intensive nature of traditional survey research. As a result, in recent years, social media monitoring tools and platforms have emerged to address the need for customer listening methods, as well as to harness the wealth of information available online in the form of user generated content. These tools offer means for listening to the social media users, analysing and measuring their activity in relation to a brand or enterprise, process that can lead to valuable insights from the side of enterprises regarding which strategy they should employ, how customers view their services and solutions, what the enterprise should expect in the future or which of their

  53  

offered features are not as effective as estimated. The added value of social media monitoring is that it offers access to real customers’ opinions, complaints and questions, at real time, in a highly scalable way. Moreover, another advantage is given by the speed at which one can investigate a topic of interest, which greatly exceeds that of a traditional survey approach. There is no longer the need for sample identification, question construction, contact attempts, and data collection prior to the analysis - social media monitoring tools only require access to the online comments and mentions posted by customers. These advantages are sustained by a press release on a yet unpublished study of Alterian and Microsoft, which claims to prove that social media monitoring is more precise, faster and more economical [5] than traditional expert panel analysis. For the specific reviewed example this statement is certainly true, but cannot be generalized. In the current study, we provide a list of 10 major monitoring tools and platforms available commercially: Alterian-SM2, Brandwatch, Converseon, Cymfony- Maestro, evolve24-Mirror, Meltwater-Buzz, NM Incite-My BuzzMetrics, Radian6, Sysomos and Visible Technologies-Visible Intelligence. Our list is not exhaustive since it has been composed using the criteria presented below. The intended use of the list is to provide an overview of the presented monitoring tools and platforms as well as to offer insights on the technology employed, on the basic features they provide as well as their limitations. The vendor selection is aligned to the following guidelines:

• Offer products that scale across multiple business functions, e.g. marketing measurement, market research, customer support, crisis identification, and so on [8];

• Offer a combination of software and services, i.e. provide proprietary dash- boards, crawlers and sentiment analysis engines [8];

• Considerable presence in the market, i.e. they are amongst the most relevant vendors on the market [8];

• Availability of technical information that could be gathered from the tools’ official websites and online reviews.

Considering the information offered on the official websites of the tools, we have selected tools that are in accordance with the first two criteria mentioned. Regarding the third criterion, we have chosen tools that have been mentioned by reports created by major technology and market research enterprises, such as Forrester Research [8], and agency-client relationship experts, such as RSW/US [15]. Moreover, it should be mentioned that our assessment criteria has been developed on the information provided on the official websites. Furthermore, we have selected only commercially available tools, since free tools have the tendency to either offer limited support or non-customizable options. However, we have included in the discussion section a short review of such free social media monitoring tools. Considering the large number of social media monitoring tools available, enterprises need an evaluation criterion to help them select the tool that is compliant with their needs and goals. The current study contributes by offering an insight of the features provided by ten of the most important social media monitoring tools available commercially, as well as provides a series of criteria for future evaluations of such tools. The remainder of the paper is structured as follows: Section 2 presents the application fields and motivation for using social media monitoring tools. Section 3 focuses on the proposed criteria for the evaluation framework split in three main categories: concepts used, technologies employed and the extent of the user interface. In Section 4 we present short descriptions for the tools chosen in the study. Section 5 focuses on the discussion and actual comparison of the tools based on the criteria chosen. The last section presents the conclusions and contribution of the present study.

9 Application fields

Posts on social media sites, in blogs and forums are relevant to many different stakeholders in an enterprise. For business leaders they provide insight in the overall online reputation of the own brands, competitors, products and services, thereby help define future strategies.

  54  

Marketing can use the insight to control feedback on campaigns. Additionally, service may identify current pain points and requests, whilst product and innovation management could derive new ideas and so forth. In a market study on social media monitoring tools of Fraunhofer IAO [1] the following application fields have been specified:

• reputation management; • event detection, issue and crisis management; • competitor analysis; • trend and market research plus campaign monitoring; • influencer detection and customer relationship management; • product and innovation management.

The keyword-sets that trigger a monitoring in each of these application fields are dissimilar and so are the positions in the enterprise that will utilize the results of a respective monitoring. Therefore, identifying the information needs must be the first step in setting up a social media monitoring project. The innovation mining process of identifying information need(s), collecting information, processing results, analyzing and interpreting and disseminating and acting is being presented by Finzen and Kintz (2011) [2]. This process has been used as a general reference for building a social media monitoring framework by Kasper and Kett (2011) [3], which includes application fields, themes, functions, activities and roles in the enterprise. The importance of social media management is further discussed in the conclusion.  

10 Evaluation framework

Currently there are more than 200 available social media monitoring tools on the market, thus making an educated choice about which tool to use has become increasingly difficult. Moreover, creating an evaluation framework for such tools has been a challenge for many reviewers and market research enterprises. For instance, Forrester [8] assesses tools based on three criteria: current offering (services and features offered), strategy (how they address enterprise-level needs) and market presence. However, the proposed criteria are insufficient. We have chosen to create a more detailed framework that focuses on the basic features of a social media monitoring tools, as well as on the technology and user interface features.

This section is split in three categories that address the following issues: the main concepts related to social media monitoring (analysis, insights, engagement, workflow management and influence); the technology used by the tools and the most important aspects related to the user interface.  

10.1 Concepts

“Concepts” refers to the elements that define a social media monitoring tool: ability to gather data and analyze it in a meaningful way to the client (illustrated by the Analysis concept), features that would enable the client to reach out to the customers (Engagement) and determine the influencers (Influence), as well as tools that allow different members of the enterprise to communicate with the tool (Workflow management). The current subsection will shortly describe these concepts and their importance. From the evaluation view, we have to take in consideration the presence of the various concepts in the tools.

Analysis The social media monitoring tool selected should be able to gather data from many sources and in different forms (e.g. posts, pictures, videos) and to establish a listening grid to capture such data. Having established a listening grid that captures data and posts around the topics the user is interested in, the next step is to analyze the data and produce actionable reports and insights for the user of the tool. The analysis is of particular importance as it encompasses the

  55  

methods used to both filter the gathered data of unwanted information (e.g. spam, duplicates) and to process it (e.g. determine the language or sentiment) in a way that is meaningful for the enterprise. The analysis should provide:

• Brand monitoring and reputation management • Consumer segmentation, customer insight and market research • Identify specific conversations to join • Gather information about competitors • Support product and service development

Engagement The engagement concept refers to the ability of the tool to support reaction with the social media posts. Many tools today offer the integrated possibility to reply to posts and follow up to any mention, complaint or question that is needed or has some opportunities.

Workflow management  Workflow refers to the process of assigning, tracking and responding to social media streams, usually in a team environment in order to prevent double responses and missed opportunities. It is crucial for an enterprise tool to promote team productivity through collaboration.

Influence  Influence refers to the ability to affect other people’s thoughts, perceptions or behaviors. In the context of social media, influence refers to those posts that have an impact on people. Although influence can be neutral, positive and negative, it is important only when it has an impact on the client’s enterprise. Influence can be attributed to a single or more individuals (called influencer), websites, specific posts or comments. Enterprises must determine who is creating the posts, as well as how many people are reading. A variety of factors, from topic relevance and reach (audience), to credibility (popularity and perceived expertise of the influencer versus his potential for bias) shape the influence concept. Social media monitoring tools should be able to determine who the influencers and brand advocates are, as well as the main detractors.

10.2 Technology  

In this subsection, we describe the technological features social media monitoring tools should provide in order to determine the extent of the effect of social media posts on the client’s enterprise, in relation to the concepts presented in Section 3.1. Additionally, these features are the building blocks required to collect data, perform the analysis and return valuable insight to the client.

Listening Grid adjustment The listening grid focuses on three main aspects: (1) the channels that are monitored (e.g. blogs and micro-blogs, social networks, video and image websites, etc.); (2) which countries and languages the tools provide support for; and (3) the topics relevant to the enterprise. Additionally, the listening grid should send alerts to inform clients (e.g. when post volume increases over a defined threshold or sentiment becomes very negative).

Near real-time processing  It is crucial for enterprises to follow up potential customers or customers’ com- plaints, questions and thoughts well in time. Therefore, the monitoring tool should provide actual data in near real-time.

  56  

Integration with 3rd party applications (API) In general, the various departments of an enterprise lean on a countable number of tools and applications. Thus, the social media monitoring tool should provide an API solution in order to make feasible the integration of the social media monitoring with other tools (e.g. customer relationship management tools).

Sentiment analysis The effort of finding valuable information in user-generated data is called opinion mining. Sentiments are determined using elements of computational linguistics, text analytics, and machine learning elements, such as latent semantic analysis, support vector machines, Natural Language Processing. Since opinion mining is a broad term, most monitoring tools have concentrated their efforts on sentiment analysis, whose main purpose is measuring the attitude, opinion, emotional state, or intended emotional communication of a speaker or writer. A sentiment score can be extremely useful in evaluating a large data set of social brand mentions, as well as allow enterprises to filter content based on positive or negative comments, thus isolating the themes or issues that have determined the developed sentiment. The major method of extracting sentiment from user generated content is Natural Language Processing (NPL). Sometimes called text analytics, data mining or computational linguistics, NPL refers to the computerized process of automatically analyzing the meaning of human language. Most current tools attempt to assign sentiment to a post automatically. Although automated sentiment technology cannot reach the quality of a human annotator, they offer comparable results to humans in center real-world scenarios. Moreover, the automatic techniques are tireless, fast, consistent (they do not make random errors), and can be improved over time [18].

Historical data Access to previously captured data is required in order to compare the current metrics and reports related to the monitored topic with any previous state of it. It is necessary to understand the improvement of a strategy in the long-run and through the years.

10.3 User Interface

Beside the accurate production of insights described in the previous subsections, the client should be able to visualize the results of the data gathering and analysis stages in a clear and concise way. Thus, the current section focuses on the user interface and the features it must possess in order to help enterprises understand their social media presence.

Dashboard In information technology, a dashboard is a user interface that organizes and presents information in a manner that is easy to read and use. To some extent, most tools’ graphical user interface resembles a dashboard. For certain tools presented in the current study, the developers consciously employ this metaphor so that the user instantly recognizes the similarity between the tool’s user interface and an automobile's dashboard. Moreover, some tools refer to their user interface as dashboards since they aim to integrate information from multiple components into a unified display. It allows users to listen, monitor and report on the conversations they are following in a quick and easy manner. These tools allow advanced configuration options for filtering language, region, media type, or organize the results found. Additionally, the dashboard offers users graphical representation of the raw data in the form of charts, listings, and historical graphing of queries and phrases. Social media monitoring tools should provide a dashboard that can be customized to the needs of the client and that includes a wide range of visualization tools. Moreover, users should be able to archive content and conversation with notes and tags, and track a sufficient number of keywords and phrases.

  57  

Export results In order to comply with their customers’ needs, some social media monitoring tools developers enable users to download the results of their tool’s analysis in different formats such as excel workbook or CSV format. In some cases, the users can create workspaces based on their preferences and download the reports.

11 Tools

This section gives a brief presentation of the tools that we are discussing in section 5 based on the evaluation framework. It is worth mentioning that the set of tools is not an exhaustive list of the available tools on the market.

11.1 Alterian - SM2

Market presence Launched in 2007. Alterian78 offers the SM2 tool, which is a business intelligence product that provides visibility into social media. In particular, it is a social monitoring and analysis tool which integrates with the other marketing solutions of the Alterian toolkit.

Technology & features The tool relies on the broad coverage of the social media by using own proprietary crawlers and data aggregators. SM2 supports sentiment analysis by utilizing a proprietary set of technologies including word parsing, weighting, proximity and Natural Language Processing. The user of SM2 has access to a comprehensive set of tools, reports and metrics that allows him/her to track and analyze conversations regarding the topic of his/her interest. Alterian users are encouraged to merge social media data with other sources of data to determine the reach, sentiment, and longevity of multi-channel campaigns. Although Alterian is one of the most popular brands on the market, the SM2 product mostly offers standard services only: the dashboard is not easily customized by the user (the infographs are basic), it offers access to an extensive historical data, performs automated sentiment analysis and filters for relevance. Moreover, [17] describes it as a mundane, expensive option, that does not offer anything new and exciting to the analytical process.

11.2 Brandwatch

Market presence Launched in August 2007. Brandwatch79 is a social media monitoring tool which focuses on gathering, “cleaning”, analyzing and presenting data. The application enables users to add their own filters of country, source, type, credibility and sentiment to analyze the data and allow the user to focus on the most relevant insights.

Technology & features The application monitors social media in four stages: (1) gathering data, (2) cleaning data, (3) analyzing and (4) presenting data. In the first stage (data gathering stage), the Brandwatch crawler operates in near-real time, gathering data based on the user’s search query from social networks, blogs and micro-blogging sites (e.g. Twitter), news services (international, national and regional), video sites, image sites, discussion forums, and corporate sites. The responses are provided in a broad data base. The data is filtered in the second stage, where irrelevant and outdated posts, as well as advertising and spam are eliminated. At this step, a Natural Language Processing algorithm notes the language used for the data, allowing the user to

                                                                                                               78 http://www.alterian.com/ 79 http://www.brandwatch.com/

  58  

filter the results by language. In the third stage, Brandwatch runs the remaining data through a five point analysis process consisting of language detection, title and content extraction, query matching, sentiment analysis and recurring phrase identifications. In the final stage the clients can use the online dashboards to create workspaces (custom reports) based on their preferences and download reports in excel workbook or CSV format. Users can create and save an unlimited number of workspaces. The dashboard does not require any software to be installed, except for a browser. Moreover, the developers provide an Application Programming Interface (API) for clients who wish to integrate Brandwatch data in their own system. Both dashboard and API function by making queries to the developer’s data-center, where data is stored using a large distributed, redundant collection of servers, guaranteeing availability and performance.

On the other hand, the application’s limitations80 are the accuracy of sentiment classification and spam filtering. However, Brandwatch states that the results can be improved using human intervention in correcting sentiment and filtering spam.

11.3 Converseon

Market presence Converseon81 offers tailor made solutions in the field of social media monitoring. Converseon was founded in 2001 as a social media agency.

Technology & features The Converseon social media monitoring toolkit utilizes the concepts of listening by mining relevant data from the social media sphere, organizing the social media campaigns of the organization and shaping its strategy in the market. Also, the toolkit supports the generation of reports and metrics regarding the performance of an organization in the market. Converseon combines technology with human analysis to obtain highly effective data quality, leading custom reports, and strong functionality [8]. However, the customized approach hurts long-term strategy, increases overall costs and slows down the pace at which customers can act on social media data [8].

11.4 Cymfony - Maestro

Market presence Cymfony Maestro82 is a third generation social media monitoring tool which gives clients (near) real time access to a comprehensive and custom built archive of traditional and social media.

Technology & features The listening and influence platform, Maestro, integrates distinctive technology with input from expert analysts to identify people, issues and trends that may impact a business. The analysis is performed in six steps: (1) gathering data; (2) refine the data to fit the customer; (3) automatic translation to English; (4) filter for spam and duplicates; (5) add value to the data (e.g. impressions, influence); and (6) Natural Language Processing for brand adds sentiment and tags of categories.

The tool acquires content in any language and is automatically translated to English. The results are presented in both the original language and English. The Maestro’s API enables the deployment of Cymfony widgets to a corporate portal, employee dashboard and other                                                                                                                80 http://www.monitoring-social-media.com/product-review-using-brandwatch-forsocial- media-monitoring 81 http://converseon.com/ 82 http://www.cymfony.com/solutions/maestro

  59  

web-based or desktop application. Additionally, Cymfony offers role-based security, a feature that is configurable based on the client’s requirements and provides a finer-grained access to the underlying data set and Maestro functionality. Cymfony Maestro provides an enterprise-class SaaS platform, configurable, on-demand and unlimited number of dashboards, as well as Natural Language Processing and sentiment analysis. The platform integrates technology with expert analysis (research and PR professionals), offering a range of packaged and custom service, in order to gather competitive intelligence, improve customer insight and market research and provide brand monitoring and reputation management. Cymfony Maestro offers an enterprise-class SaaS platform, as well as configurable, on-demand and unlimited number of dashboards. Moreover, brands can use the basic Maestro platform to retrieve, process and analyze data in near-real time, whilst those with additional analytical requirements can access categorized market intelligence reports: “Competitive Landscape Reports”, “Category Insights Reports”, and “Holistic Market Research Engagement Reports” [17].

11.5 evolve24 - Mirror

Market presence Evolve2483 offers a social media monitoring tool, called Mirror since 2004.

Technology & features The Mirror uses proprietary algorithms in order to help the user define the best strategies regarding the monitored data from the social networks. It captures data from different sources in near real-time, analyzes it and generates reports. It helps the user recognize complaints, questions and emerging threats due to the sentiment analysis and weighting system that it uses. Its main strength is its text processing data quality, while its limitations are its dashboard interface and the lack of tool sets for marketers [8].

11.6 Meltwater - Buzz

Market presence Meltwater Buzz84 helps businesses manage their social presence and engage with current and prospective customers since 2007.

Technology & features Buzz combines the capabilities of Meltwater’s Buzz tool with the features offered by the JitterJam product. The JitterJam product combines social media monitoring, a contact database and multichannel digital marketing platform in a single social CRM system, whilst Meltwater Buzz monitors, track and analyze user-generated content and social media presence. Engage focuses on specific individuals within the community rather than a generic topic in order to provide personal, trusted relationships with customers and stakeholders. Meltwater Buzz Engage provides a social dashboard, multi-user workflow and a personal social advisor (to help the user learn about the benefits of the tool, as well as provide guidance and support for the platform’s features). The Social Dashboard offers a “brandometer” (a graphical view of the overall sentiment), a visualization of sentiment, search results, “themes cloud” (snapshot of most frequent themes found in search results) and media spread (graphical view of the customer’s campaign search results by social channel). The theme cloud provides a snapshot of most frequent search result themes, as well as helps identify potentially damaging conversations. Additionally, the tool enables “campaign” searches (dedicated to a specific brand, market or topic) and identifies relevant conversations and key influencers. Furthermore, Meltwater Buzz helps clients determine the overall

                                                                                                               83 http://www.maritzresearch.com/solutions/social-intelligence.aspx 84 http://buzz.meltwater.com/

  60  

sentiment of social conversations about the clients’ brand and understand social trends (the volume, sentiment and media spread of conversations).  

11.7 NM Incite - My BuzzMetrics

Market presence Nielsen & McKinsey launched a joint venture in 2010, the NM Incite. NM Incite offers a leading listening platform (My BuzzMetrics). However, Nielsen is active in the domain of social media monitoring since 1997.

Technology & features My BuzzMetrics makes feasible the gathering of data from social media, as well as filtering out any noise from the monitored social conversations towards focusing on relevant topics. Moreover, the toolkit of NM Incite provides the user with reporting tools and the opportunity to react with the customers in real time via the social media profiles of the enterprise. My BuzzMetrics relies on automatic sentiment analysis and a wide range of social media sources. There is also an API available to connect any existing tools of the user to the data feeds and reports of the platform. NM Incite’s limitation consists in its poor user experience determined by the interface and functionality [8].

11.8 Radian6

Market presence Radian685 delivers one of the most popular social media monitoring tools in the market. Radian6 launched in 2006 and was acquired by SalesForce in 2011.

Technology & features It enables organizations to become Socially Engaged Enterprises by providing means for listening to social media, analyzing and measuring the raw data, producing insights based on Natural Language Processing and engaging with the streams of posts in the social media sphere (engagement console). Moreover, the users of Radian6 are able to use a summary dashboard tool in order to get in brief the status of their monitored topics. The Radian6 platform allows the administrator of an enterprise to supervise and orchestrate his/her team by exploiting the features of the workflow management console included in the engagement console. Thus, (s)he is able to assign conversations and posts to his/her colleagues in order to take care of them, specify the priority of the tasks and also classify them (e.g. question, complain etc.). Compared to the other tools presented, Radian6 dashboard offers the basic features as well as attractive infographics. Moreover, the developers provide software capable of performing CRM, which connects the dashboard with the sales databases in order to create contacts and leads from the “River of News”. Radian6’s major strengths are its comprehensive coverage of social media data, its scalability and ability to integrate with other enterprise applications. On the other hand, Radian6 offers only 30 days of immediate historical data, which is not sufficient for a brand that wishes to analyze sentiment before, during and after the implementation of a new product or service. Moreover, due to the many features and functions, a user experiences a learning curve.

11.9 Sysomos

Market presence Sysomos86 provides tools for monitoring social media conversations and themes, identify key influencers and gather insight and intelligence to help shape the business decisions and strategies of the client’s enterprise.                                                                                                                85 http://www.radian6.com/

  61  

Technology & features Sysomos offers two tools for social media monitoring: Media Analysis Plat- form (MAP) and Heatbeat. MAP is a feature-rich service that mines and analyzes social media and user-generated content. MAP provides a comprehensive and spam-free database with content gathered constantly by actively indexing blogs, social networks, Twitter, YouTube, wikis, messages boards, video sharing sites, and news sources. Additionally the tool offers an automated sentiment engine, measurable metrics, key influencer identification, competitive analysis, global and multi-language support, a full-featured engagement workflow and detailed demographics of the results. Heartbeat offers cost effective and near real-time social media monitoring and measurement, providing engagement capabilities for professionals, brand managers and customer support groups delivered using a variety of user-friendly and intuitive graphics. In essence, Heartbeat provides a subset of MAPs features, with a strong focus on enabling companies to track social media, organize conversations, manage workflows, facilitate collaboration, and provide ways to engage with influencers. The underlying content aggregation and analysis engine commercialized by Sysomos has been created as part of the BlogScope project, a free search service for the blogosphere developed as part of a research project between Sysomos Inc. and the University of Toronto.

11.10 Visible Technologies - Visible Intelligence

Market presence Visible Intelligence87, a product of Visible Technologies, enables users to monitor, analyze and actively engage in social media conversation using a single environment. The product was launched in 2005.

Technology & features Visible Intelligence has been designed for marketers, research groups, agencies and any other enterprise department that wants to monitor and analyze social media trends to maximize ROI. The tool is designed to handle substantial amounts of information, manage alliances, workflows, assimilate and adjust to the client’s processes and systems. Visible monitors the brand and manages the reputation of the client’s enterprise, providing consumer segmentation, crisis communication, support of customer services efforts and support product/service development. Moreover, the tool gathers competitive intelligence and tracks social media initiative in order to identify brand advocates (influencers) or specific conversations to join. Visible provides Web crawling technology, API integrations, text mining and filtering, as well as Natural Language Processing and text analysis (at the topic level). Furthermore, the tool offers both human analysis and manual analysis’ refinement by client to help the client understand the landscape and determine which intelligence to use. The primary mode of delivery for the analysis results is browser-based dashboard and self-service tools. On the other hand, Visible Technologies does not approach the subject of context or limitations of NLP engines. Moreover, [17] claims that apart from positive client testimonials, there is little evidence that Visible Intelligence is superior to cheaper competitor tools.

12 Discussion

To evaluate the products against our set of criteria, we have gathered details based on a combination of information offered by official websites for the tools or vendors, white papers and external product reviews. We have set weights to reflect our analysis, as well as score the tools on a clearly defined scale, following the criteria proposed in Section 3. The weights offered are intended only as starting point, since our analysis did not include a test on the actual tools and on the underlying technologies they employ. The evaluation framework,

                                                                                                                                                                                                                                                                                                                             86 http://www.sysomos.com/ 87 http://www.visibletechnologies.com/

  62  

described in the third section, provides a set of criteria that the authors of this paper find important and useful for the comparison of social media monitoring tools in the market. The purpose of this evaluation is to help potential users of social media monitoring tools make the right choice. Our purpose is not to provide an exhaustive list of tools and a detailed survey on their features in order to conclude which of them addresses the criteria in the best way. Thus, Table 3 demonstrates a way to evaluate a set of tools towards the evaluation framework that this paper introduces. Regarding the notations used in the table: we use the check mark () in case the criterion is fulfilled and the () in case it is not supported. For the criteria that we have not sufficient information, we leave the related space blank.

Tool name Ana

lysi

s En

gage

men

t W

orkf

low

Man

agem

ent

Soci

al P

rofil

es

List

enin

g G

rid

Nea

r rea

l-tim

e pr

oces

sing

A

PI

Sent

imen

t Ana

lysi

s H

isto

rical

dat

a D

ashb

oard

Ex

port

resu

lts

Alterian SM2 Brandwatch Converseon Cymfony Maestro evolve24  Mirror Meltwater  Buzz NM  Incite  My  BuzzMetrics Radian6                        Sysomos   Visible  Technologies  Intelligence  

Table 3. Evaluation of tools on the criteria

Besides the users’ time, tool-costs of up to several thousand Euros a month [1] arise when introducing professional social media monitoring. Using free services like the ones shown in Table 4 are a cost-efficient alternative especially when starting exploring this field, but the following aspects need to be considered: (1) Free tools are also free of service. There is no contact person answering questions on functions, underlying methodology and there is no guarantee concerning the availability of the service. (2) Functions are often limited to quantitative/statistical reports. Complex analysis, for example the automated sentiment detection, is not available for languages other than English in free tools. (3) Many of the free tools are point solutions considering few or only one platform like Twitter. Services that claim searching the entire web do not reveal which sources are really included. To get a comprehensive overview several free services must be combined. (4) Results of free tools have to be saved and archived in user-defined structures and formats for example for adding own comments and especially when consolidating data from more than one free tool. (5) Workflow-functions for example for forwarding findings to colleagues and tracking the processing of such a finding are usually not available.

Addict-o-matic www.addictomatic.com

Boardreader www.boardreader.com

Google Alerts www.google.com/alerts

HyperAlerts www.hyperalters.no

  63  

Klout www.klout.com/home

Netvibes www.netvibes.com

Twazzup www.twazzup.com

WhosTalkin www.whostalkin.com

Yahoo Pipes pipes.yahoo.com

Table 4. Some free social media monitoring tools  

13 Conclusion

In the introduction we have discussed how social media monitoring tools enable enterprises and other institution to gather direct and up-to-date data which can provide valuable insights for their business in various application fields. To support choosing such a tool studies on social media monitoring tools have been presented for example by Plum [6], Gilliat [7], Kasper [1] and Forrester Research [8]; however, they are quickly outdated due to the rapid development of the market: new functionalities, takeovers and the appearance of new players make it difficult to solely rely on such studies. A promising approach is to make information on social media monitoring tools instantly available and editable through web-platforms like www.medienbewachen.de or www.somemo.at. However, since the information is entered by the vendors, this concept might lack objectivity. To support the process of choosing a social media monitoring tool we have introduced an evaluation framework and exemplarily applied this on a set of professional tools. The framework provides a basic structure which can be further detailed for example by specifying data sources and regional context of this data. Social media offers new opportunities for enterprises, both in monitoring conversations and in actively participating and providing content on social media platforms. Social media monitoring tools support these activities; however, an enterprise also needs social media management, that means the definition of strategies, roles and processes in this new field. Due to an incident known as “Dell Hell” [9] the enterprise Dell has started early to set up structures to deal with the new communication paradigm where organizations no longer only push information through mass media, but engage in conversations with the customer. A social media listening command center has been installed to address social media conversations [10] and furthermore all employees are trained and empowered to speak on behalf of the enterprise. Enterprises must realize, that social media is not only a topic for corporate communications or marketing, but needs to be addressed cross-departmental. The Social Media Governance Study 2011 [11] shows that “greater experience of organizations results in the incorporation of more areas”. Choosing the right tools and instantiating an appropriate organization for one’s strategy are key factors for businesses to benefit from social media.

References

[1] Kasper, H., Dausinger, M., Kett, H., Renner, T.: Marktstudie Social Media Monitoring Tools - IT-Lösungen zur Beobachtung und Analyse unternehmensstrategisch relevanter Informationen im Internet. Fraunhofer Verlag, Stuttgart (2010)

[2] Finzen, J., & Kintz, M.: Innovation Mining - Supporting Web Mining in Early Innovation Phases. Proceedings of the 7th WEBIST conference (2011)

[3] Kasper, H., Kett, H.: Social Media Monitoring-Tools. In T. Schwarz, ed. Leitfaden Online-Marketing. Das Wissen der Branche. Waghäusel: Marketing-Börse, pages 662-669 (2011)

[4] Kasper, H., Dukino, C., Kett, H.: Marketing und Vertrieb für Cloud-Angebote. In D. Spath, N. Weiner, T. Renner, A. Weisbecker: Neue Geschäftsmodelle für die Cloud entwickeln - Methoden, Modelle und Erfahrungen für Software-as-a-Service im Unternehmen. Fraunhofer Verlag, Stuttgart (2012)

[5] Alterian press release retrieved from http://www.fair-news.de/print/453705/ (2011) [6] Plum, A.: Ansätze, Methoden und Technologien des Web-Monitorings - ein systematischer Vergleich. In:

Brauckmann, P. (ed.): Web-Monitoring - An approach for evaluation of social media monitoring tools 17

  64  

Gewinnung und Analyse von Daten über das Kommunikationsverhalten im Internet. Pp. 21-46, UVK, Konstanz (2010)

[7] Gilliatt, N.: Social Media Analysis Platforms for Workgroups. Social Target LCC (2010) [8] Hofer-Shall, Z., Murphy, E., Grant, M., Vittal, S.: The Forrester Wave: Listening Platforms, Q3 2010. [9] Fung, M. L.: You Can Learn Fraom “Dell Hell.” Dell Did. Retrieved from

http://www.customerthink.com/article/you can learn dell hell dell did (2008) [10] Swallow, E.: Dell To Launch Social Media Listening Command Center. Retrieved from

http://mashable.com/2010/12/08/dell-social-listening-center/ (2010) [11] Fink, S., Zerfa, A., Linke, A.: Social Media Governance 2011 - Expertise Levels, Structures and Strategies

of Companies, Governmental Institutions and Non-Profit Organizations communicating on the Social Web. University of Leizpig / Fink & Fuchs Public Relations AG, Leipzig / Wiesbaden (2011). Retrieved from www.socialmediagovernance.eu

[12] Perdue, D. J.: Social Media Marketing: Gaining a Competitive Advantage by Reaching the Masses, (2010) Retrieved from http://digitalcommons.liberty.edu/honors/

[13] VanBoskirk, S.: Interactive marketing nears $55 billion; advertising overall delines. (2009) Message posted to http://blogs.forrester.com/

[14] Murphy, J., Kim, A.E., Hagood , H., Richards, A.K., Augustine, C.B., Kroutil, L.A.: Twitter Feeds and Google Search Query Surveillance: Can They Supplement Survey Data Collection?, (2011) Retrieved from http://www.rti.org/publications/abstract.cfm?pubid=17912

[15] RSW/US & Web Liquid: Marketers & Social Media Monitoring Survey (2011) [16] Power Reviews: Social Commerce Stats, (2011) Retrieved from

http://www.powerreviews.com/resources/social-commerce-stats [17] Boardtracker: The Wilderness of Mirrors a review of the top 50 social media analysis tools and what they

don’t tell you, (2011) Retrieved from http://socialmediatoolkit.wikispaces.com/Monitoring+and+Tracking [18] Kmetz, J.: Measuring Social Sentiment: Assessing and Scoring Opinion in

Social Media, Retrieved from: http://www.visibletechnologies.com/resources/white-papers/measuringsentiment/