Modeling User Interactions in Online Social Networks (2009)
Asian Workshop of ASWC 2009
Social Web and Interoperability Dec. 7th , Shanghai, China Modeling User Interactions in Online Social Networks to solve real problems Seokchan (Channy) Yun and Hong-Gee Kim Biomedical Knowledge Engineering Laboratory Seoul National University, Korea
Agenda • Introduction – Some
approaches for Social Semantic Web • Challenges – Finding the definition of online friends and interaction between users • Survey of social interaction in real SNS – Twitter and Me2day • Result and Discussion • Conclusion and Future plan
• New opportunities for social
science – Explicit and implicit social network information – Large scale and dynamic data sets – Different modalities (profiles, email, IM, Twitter…) • Challenges – Friend on the Web = Friend in reality? – Heterogeneity and quality of data – Time and space complexity – Ethical and legal challenges – Complex interaction = Centrality in reality?
History • First Mover –
Classmates.com, Match.com and sixdegree.com – Friendster and Orkut • Majority – Myspace – Facebook – Linkedin – Twitter
How succeed? • Allows a
user to create and maintain an online network of close friends or business associates for social and professional reasons: – Friendships and relationships – Offline meetings – Curiosity about others – Business opportunities – Job hunting • Allows a user to share interests based on object-centered sociality with meaning – Sharing photo, video and bookmark – Life streaming over SNS – Broadcasting and publishing of my own content
Tripartite Social Ontology (Peter Mica)
• A graph model of ontologies based on tripartite graphs of actors, concepts and instances – Actors: users – Concepts: tags – Instances: objects • Emergent semantics – General idea: observe semantics in the way agents interact (use concepts) • Bottom-up ontologies • Semantics = syntax + statistics
Limitations • FOAF – Only
focusing on ONE PERSON • SIOC – Only focusing on relationship with site (forum), contents and person. • Tripartite Social Ontology – Too high abstraction level to be implemented • Online Presence Project – Only focusing “Presence” not to be interested in “Activity • Activity streams – Only description for Person / Verb / Object • SemSNI – Only can be applied in specific domain if you have all data
What’s real problems? • Twitter
– There are many spammers and followers. – Whom I should follow? Who is expert? • me2DAY (or Facebook) – There are many friends – Who disconnected in my friendship? • Flickr – There are many photos. – What’s good photos enjoying with friend? • RateMDs – There are many doctors. – What’s good doctors recommended by friends?
1. What’s definition of Online
Friend? http://answers.polldaddy.com/poll/1230119/?view=results Well-known Friends 9% Colleagues 7% Known Meet once in offline 25% Knowing only name 12% Famous person 3% Unknown friend of friends 13% Unknown Everyone who requests 32% Online Friend != Real FOAF’s knows is not knowing!
Challenges • Online friends and
interaction are not real because there are no limits of time and space. • It’s hard to find degree of user relationship. – Coupling-decoupling between users (high vs. weak) by time change • We have to consider the difference of each online interaction to measure proper centrality and betweenness.
Approach • Sample data analysis
of Me2day and Twitter – Developing Twitter application: Twi2me • Twi2me helps for user to post Tweets to me2day in real-time. – Me2day: gathering interaction on purpose of research of 32,200 accounts from January to October, 2009 – Twitter: gathering interaction 1,120 users on time of Oct. 12th , 2009 • Measuring differences of social interaction – Classification of user-interaction – Analysis of interaction statistics
Results : me2day Kinds of
interaction Numbers Reply 2,074,284 Comments between users Metoo 451,260 Similar with Retweets Private Messages 31,915 Similar with Direct Messages SMS 30,000 Short message by phone Gift 3,590 Sharing items in SNS
Result: Twitter Total Messages 10000
1000 Reply 100 ReTw eet 10 DM 1 10 100 1000 10000 Total Followers • Surveyed by total 1,120 Twitter users in Korea – Reply interaction is growing along with followers. – ReTweet and Direct Message are less than reply
Suggestion: Interaction Index Nb. Of
Interaction Interaction Index Impact of Interaction Reply 2,074,284 0.8006 1.00 Metoo 451,260 0.1742 4.60 Private Messages 31,915 0.0123 64.99 SMS 30,000 0.0116 69.14 Gift 3,590 0.0014 577.79 Total 2,591,049 1.0000 Comparing with Reply • If the interaction index is “1”, it’s general relationship. • Ratio compared with interaction index between user A and B is strength of betweenness.
Discussion • Q: Interaction depends
on user experience? – User tends to do easy interactive method. – ReTweet is harder than reply in Twitter. • A: User does emotional interaction. – For example, agreement and consensus • Metoo is easier than comment in me2day • ReTweet is easier than direct message in Twitter – But, • Nb. of comment > Nb. Of metoo • Nb. of direct message == Nb. of ReTweet (Information distribution)
Conclusion • Difference of strength
in user interaction – Twitter: • Reply < ReTweet < Direct Message < SMS – me2Day • Comment < metoo < Private Messages < SMS < Gift • Measuring strength of user relationship – Modeling of user degree – Measuring Interaction Impact – Similarity formula (A,B) • Solving problem after integration data
Future Plan • Social web
evolves direct sharing and broadcasting instead of document link based distribution and knowledge discovering. – Social Interaction is more important in social networks. – FriendFeed, Facebook life streaming, Twitter • Need to represent “Degree between people” – Writing simple ontology represents interaction • Channy replies Hong-Gee (What) (When) in Facebook • John retweets Channy (What) (When) in Twitter – Extending ActiveStreams or SemSNI
• Who disconnected in my
friendship on me2DAY? – Gathering me2day activities – Measuring interaction factor and coupling degree • Distance = # of interaction/ time interval • Priority = normalized value for each interactions – Evaluation with user’s reaction for alert • “Why don’t you contact this person because it’s long time not to contact by you?” • Whom I should follow? Who is expert in Twitter? – Gathering twitter activities – Measuring interaction factor and coupling-degree – Evaluation with user’s reaction for recommendation