Chapter 8 A Semantic Web Primer1
Chapter 8Conclusion and Outlook
Grigoris Antoniou
Frank van Harmelen
Chapter 8 A Semantic Web Primer2
Lecture Outline
1. Which Semantic Web?
2. Four Popular Fallacies
3. Current Status
4. Selected Key Research Challenges
Interpretation 1: The Semantic Web as the Web of Data
The main aim of the Semantic Web is to enable the integration of structured and semistructured data sources over the Web
The main recipe is to expose datasets on the Web in RDF format and to use RDF Schema to express the intended semantics of these datasets
A typical use is the combination of geodata with a set of consumer ratings for restaurants in order to provide an enriched information source
Chapter 8 A Semantic Web Primer3
Interpretation 2: The Semantic Web as Enrichment of the Current Web
The aim of the Semantic Web is to improve the current World Wide Web
Typical uses are :– Improved search engines– Dynamic personalization of Web sites– Semantic enrichment of existing Web pages
Chapter 8 A Semantic Web Primer4
Interpretation 2: The Semantic Web as Enrichment of the Current Web (2)
The sources of the required semantic metadata are mostly claimed to be automated sources– Concept extraction – Named-entity recognition– Automatic classification
More recently the insight is gaining ground that the required semantic markup can also be produced by social mechanisms in communities that provide large-scale human-produced markup
Chapter 8 A Semantic Web Primer5
Chapter 8 A Semantic Web Primer6
Lecture Outline
1. Which Semantic Web?
2. Four Popular Fallacies
3. Current Status
4. Selected Key Research Challenges
1. The semantic Web tries to enforce meaning from the top
This fallacy claims that the Semantic Web enforces meaning on users through its standards OWL and RDFS
The only meaning that OWL and RDFS enforce is the meaning of the connectives in a language in which users can express their own meanings
Users are free to choose their own vocabularies and to describe whatever domains the choose
Chapter 8 A Semantic Web Primer7
1. The semantic Web tries to enforce meaning from the top (2)
The situation is comparable to HTML:– HTML does not enforce the lay-out of web-pages
“from the top”– All HTML enforces is the language that people
can use to describe their own lay-out – HTML has shown that such an agreement on the
use of a standardized language is a necessary ingredient for world-wide interoperability
Chapter 8 A Semantic Web Primer8
2. Everybody must subscribe to a single predefined meaning for the terms they use
The meaning of terms cannot be predefined for global use in addition meaning is fluid and contextual
The motto of the Semantic Web is not the enforcement of a single ontology but rather “let a thousand ontologies blossom”
That is the reason that the construction of mappings between ontologies is such a core topic in the Semantic Web community
Such mappings are expected to be partial, imperfect and context-dependent
Chapter 8 A Semantic Web Primer9
3. Users must understand the complicated details of formalized knowledge representation
Some of the core technology of the Semantic Web relies on intricate details of formalized knowledge representation
The semantics of RDF Schema and OWL and the layering of the subspecies of OWL are difficult formal matters
The design of good ontologies is a specialized area of Knowledge Engineering
Chapter 8 A Semantic Web Primer10
3. Users must understand the complicated details of formalized knowledge representation (2)
For most users of Semantic Web applications, such details will remain entirely behind the scene
Navigation or personalization engines can be powered by underlying ontologies, expressed in RDF Schema or OWL, without users ever being confronted with the ontologies, let alone their representation languages
Chapter 8 A Semantic Web Primer11
4. The semantic Web requires the manual markup of all existing Web pages
It is hard enough for most Web site owners to maintain the human-readable content of their sites
They will certainly not maintain parallel machine-accessible versions of the same information in RDF or OWL
If that were necessary, it would indeed spell bad news for the Semantic Web
Instead, Semantic Web application rely on large-scale automation for the extraction of such semantic markup from the sources themselves
Chapter 8 A Semantic Web Primer12
Chapter 8 A Semantic Web Primer13
Lecture Outline
1. Which Semantic Web?
2. Four Popular Fallacies
3. Current Status
4. Selected Key Research Challenges
Four Main Questions
Where do the metadata come from? Where do the ontologies come from? What should be done with the many
ontologies? Where’s the "Web " in Semantic Web?
Chapter 8 A Semantic Web Primer14
Question 1: Where do the metadata come from?
Much of the semantic metadata come from Natural Language Processing and Machine Learning technology
It is now possible with off-the-shelf technology to produce semantic markup for very large corpuses of Web pages by annotating them with terms from very large ontologies with sufficient precision and recall to drive semantic navigation interfaces
Chapter 8 A Semantic Web Primer15
Question 1: Where do the metadata come from? (2)
More recent is the capability of social communities to provide large amounts of human-generated markup
Millions of images with hundreds of million of manually provided metadata tags are found on some of the most popular Web 2.0 sites
Chapter 8 A Semantic Web Primer16
Question 2: Where do the ontologies come from?
The term ontology as used by the Semantic Web community now covers a wide array of semantic structures, from lightweight hierarchies such as MeSH to heavily axiomatized ontologies such as GALEN
The world is full of such “ontologies”:– Companies have product catalogs– Organizations have internal glossaries– Scientific communities have their public metadata
schemata
Chapter 8 A Semantic Web Primer17
Question 2: Where do the ontologies come from? (2)
These have been constructed for other purposes, most often predating Semantic Web
There are also significant advances in the area of ontology learning, although results there remain mixed
Obtaining the concepts of an ontology is feasible given the appropriate circumstances, but placing them in the appropriate hierarchy with the right mutual relationships remains a topic of active research
Chapter 8 A Semantic Web Primer18
Question 3: What should be done with the many ontologies?
The Semantic Web crucially relies on the possibility of integrating multiple ontologies
This is known as the problem of ontology alignment, ontology mapping, or ontology integration
Chapter 8 A Semantic Web Primer19
Question 3: What should be done with the many ontologies? (2)
A wide array of techniques is deployed for solving this problem – Ontology-mapping techniques based on natural
language technology– Machine learning – Theorem proving– Graph theory – Statistics
Chapter 8 A Semantic Web Primer20
Question 3: What should be done with the many ontologies? (3)
Although there are encouraging results, this problem is by no means solved
Automatically obtained results are not yet good enough in terms of recall and precision to drive many of the intended Semantic Web use cases
Ontology mapping is seen by many as the Achilles’ heel of the Semantic Web
Chapter 8 A Semantic Web Primer21
Question 4: Where’s the "Web " in Semantic Web?
Semantic Web has sometimes been criticized as being too much about “semantic” and not enough about “Web”
This was perhaps true in the early days of Semantic Web development, when there was a focus on applications in rather circumscribed domains like intranets
The main advantage of company intranets is that the ontology-mapping problem can be avoided
Chapter 8 A Semantic Web Primer22
Question 4: Where’s the "Web " in Semantic Web? (2)
Recent years have seen a resurgence in the Web aspects of Semantic Web applications
A prime example is the deployment of FOAF technology, and of semantically organized P2P systems
The Web is more than just a collection of textual documents
Chapter 8 A Semantic Web Primer23
Question 4: Where’s the "Web " in Semantic Web? (3)
Nontextual media such as images and videos are an integral part of the Web
For the application of Semantic Web technology to such nontextual media we currently rely on human-generated semantic markup
Deriving annotations through intelligent content analysis in images and videos is under way
Chapter 8 A Semantic Web Primer24
Main Application Areas
Looking at industrial events, either dedicated events or co-organized with the major international scientific Semantic Web conferences, we observe that a healthy uptake of Semantic Web technologies is beginning to take shape in the following areas :– Knowledge management, mostly in intranets of large
corporations– Data integration (Boeing, Verizon, and other)– E-science, in particular the life sciences– Convergence with Semantic Grid
Chapter 8 A Semantic Web Primer25
Main Application Areas (2)
If we look at the profiles of companies active in this area, we see a transition from small start-up companies such as
– Aduna– Ontoprise – Network Inference – Top Quadrant
To large vendors such as– IBM (Snobase ontology Management System)– HP (Jena RDF platform)– Adobe (RDF-based XMP metadata framework)– Oracle (support for RDF storage and querying in their datavase
product)
Chapter 8 A Semantic Web Primer26
Main Application Areas (3)
However, there is a noticeable lack of uptake in some other areas. In particular, the promise of the Semantic Web for– Pesonalization– Large-scale semantic search (on the scale of the
World Wide Web)– Mobility and context-awareness
is largely unfulfilled
Chapter 8 A Semantic Web Primer27
Main Application Areas (4)
A difference that seems to emerge between the successful and unsuccessful application areas is that the successful are all aimed at closed communities, whereas the applications aimed at the general public are still in the laboratory phase at best
The underlying reason for this could well be the difficulty of the ontology mapping
Chapter 8 A Semantic Web Primer28
Chapter 8 A Semantic Web Primer29
Lecture Outline
1. Which Semantic Web?
2. Four Popular Fallacies
3. Current Status
4. Selected Key Research Challenges
Selected Key Research Challenges (1)
Several challenges that were outlined in 2002 article by van Harmelen have become active areas of research:– Scale inference and storage technology, now
scaling to the order of billions of RDF triples– Ontology evolution and change– Ontology mapping
Chapter 8 A Semantic Web Primer30
Selected Key Research Challenges (2)
A number of items on the research agenda, though hardly developed, have had a crucial impact on the feasibility of the Semantic Web vision:
– Interaction between machine-processable representations and the dynamics of social networks of human users
– Mechanisms to deal with trust, reputation, integrity and provenance in a semiautomated way
– Inference and query facilities that are sufficiently robust to work in the face of limited resources (computational time, network latency, memory, or storage space) and that can make intelligent trade-off decisions between resource use and output quality
Chapter 8 A Semantic Web Primer31