a diffusion mechanism for social advertising over microblogs
TRANSCRIPT
A diffusion mechanism for social advertising over microblogs
Mohammad Moein Hosseini Manesh
M.S student of Information Technology (e-commerce)
Amirkabir University of Technology
July - 2016
Online Adversiting
Targeted AdvertisingBased on user's preferenceContent-based approach
Collaborative-based approach
hard to recommend new items to users when there are no related comments or rating records
Social AdvertisingBased on social relationshipUses social relations and social influences between people to sell products or services
Indirect Advertising: Word-of-mouth / Viral marketing
Social Advertising
Goal: Influence more users with a fixed budget
How: Disseminate advertisments via social endorsers
Endorsers: Most influential users of social media
Find endorsers of your market segment
Motivate them to share your product/service
Architecture of endorser discovery engine
Preference analysis module
User preferencePre-Classiying
Targeted Marketing: Right product to right user
Advertisement fitnessEvaluating advertisement matching
Using advertisment ontology tree
Better result than tag-based approach
Fitness
Influence analysis module
Influential LeaderFreeman:Degree Centerality
Betweenness Centerality
Kiss and Bichler:Out-degree Centerality
Kwak:In-degree centerality equals to Page Rank
Our Approach:Connection degree (Freinds)
Content Popularity
Influence analysis module
Connection degree influenc:Degree centrality: the number of direct connections/links upon a node
mutual relationship (friendship)
Content popularity influencEvaluate the popularity of what a user posts
By the number of total responses and message reposts from people
||: The total number of the elements in a set
Propagation strength analysis module
Social activenessCalculate the activity intensity of a user
Higher activeness higher probability of message sharing
Propagation strength analysis module
Social similaritySimilarity between two users by implicit social structure and behaviorSocial structure Firends in common
Behavior Contents in common
Friends tend to have similar interests if people have more common friends, their interests should be more similar
Propagation strength analysis module
Social similarityCommon Friends:
Common Contents:IR approachesLanguage Model
TF-IDF
...
Propagation strength analysis module
Common Contents:
Propagation strength analysis module
Social interaction:Social interaction explicitly catches the factor of dynamic actions/interactions between people
The intimacy between two users
More interaction activities, the higher the probability of sharing information
Propagation strength analysis module
Social propagation:Evaluate the network diffusion strength of a user
Friends' social propagation capability would also affect a social endorser's propagation score recursive
How to weight?
Black-box methodArtificial neural networkcan be trained to recognize and generalize the relationship between a set of inputs and outputs by nonlinear processes
Required a training data for learning ANN.
White-box methodAnalytic hierarchy process
Weights:Preference: 0.340075768
Influence: 0.325100803
Propagation: 0.33482343
Evaluation
Plurk micro blogging
217 Start users
55% Male, 45% Female
Aged between 20 and 50
Total Users: 247,900
Three months crawling
216 advertisments samples (12 sets)
Evaluation
4 different approachesIn-degree
Ratio-degree followers - followings
Topic-influence preference + influence
Our approach
IndicatorsClick-through rate
Repost rate
Exposure rate
The Category Tree
Evaluation : CTR
Evaluation : RR
Evaluation : EX
Evaluation
4 different approachesIn-degree
Ratio-degree followers - followings
Topic-influence preference + influence
Our approach
IndicatorsClick-through rate
Repost rate
Exposure rate
Future works
User prefrence discovery improvment
Shortest path
Real semantic analysis
Using sentiment analysis
Thanks For Your Attention