1. Feedback Effects between Similarity and Social Influence in Online Communities David Crandall, Dan Cosley, Daniel Huttenlocher, Jon Kleinberg, Siddharth Suri Cornell University (KDD ’ 08) Advisor: Dr. Koh, JiaLing Speaker: Cho, ChinWei Date: 2009/4/13

2. Outline  Introduction  Analyzing Similarity over Time  Modeling the Effect of Social Interaction  Predictive value  Conclusions

3. Introduction  A fundamental open question in the analysis of social networks is to understand the interplay between similarity and social ties.  People are similar to their neighbors in a social network for two distinct reasons: Social influence People become similar to those they interact with Selection People seek out similar people to interact with

4. Introduction  Social influence can push systems toward uniformity of behavior  Selection leads to fragmentation of the community fragmentation homogeneity

5. Introduction  The two forces of social influence and selection are both seen in a wide range of social settings: People decide to adopt activities based on the activities of the people they are currently interacting with People simultaneously form new interactions as a result of their existing activities

6. Introduction  Studying these forces and their interplay has been very difficult because collecting data about an individual ’ s social network and activities over time is both expensive and error-prone.  Online communities provide an excellent opportunity to study large-scale social phenomena of this type.

7. Introduction  Together these forces shape how a community develops Important for understanding health, trajectory of community  Applications in online marketing Viral marketing relies upon social influence affecting behavior Recommender systems predict behavior based on similarity

8. Introduction  We study two questions in large online communities How do selection & social influence interact to create social networks? Can we quantify and model the way in which selection and social influence interact? How do similarity and social ties compare as predictors of future behavior?

9. Interplay between influence and similarity  How does first interaction affect similarity?  Wikipedia: a large collaborative social network users interact by posting to each others ’ user-talk pages user interests revealed by article edit patterns rich, publicly-available, fine-grained log Social influence dominates? Selection dominates? Transient effect? time similarity first interaction time similarity first interaction time similarity first interaction

10. Analyzing Similarity over Time  We focus on two types of recorded actions: Editing articles  Edits to articles are indications of interests as each article is on a particular topic.  We take the history of article edits for a user at a given time as a vector encoding that user ’ s activities  Consider the time series of these vectors as representing the evolution of users ’ interests and activities as they edit different articles. Editing the discussion page associated with a particular user  Edits to discussion pages are indications of social ties between the users who are communicating.

11. Analyzing Similarity over Time  Given a set of m possible activities  Each person v at time t has an m-dimensional vector v(t), where the i th coordinate v(t) i represents the extent to which person v is engaging in activity i.  Binary: v(t) i equal to 0 or 1 depending on whether v has ever engaged in activity i  Weighted: v(t) derived in some way from the number of times v has engaged in i  Using Cosine Similarity

12. Analyzing Similarity over Time  How does the similarity between two people vary in the time window around their first interaction with each other?

13. Analyzing Similarity over Time Slower, long-term increase after first interaction (social influence) Slower, long-term increase after first interaction (social influence) Rapid increase in similarity before first interaction (selection) Rapid increase in similarity before first interaction (selection)

14. Modeling the Effect of Social Interaction  People ’ s future behavior is strongly correlated with their past behavior, and so one possibility is that a user chooses what to do next by sampling from his or her own past activity history.  The effects of social influence can be captured by assuming that a user may also choose a next activity by sampling from the past history of his or her friends in the social network.

15. Modeling the Effect of Social Interaction  An activity is something that can be performed multiple times; Each time it is performed is called an instance of the activity.  User ’ s activity history E(u) is simply the sequence of all its instances of past activities.  Each time a user u communicates with a user v, which we will call an interaction with v.  User u ’ s interaction history N(u) is the sequence of all their past interactions.

16. Modeling the Effect of Social Interaction  At each time step, user u either, Interacts with another user, choosing someone who has engaged in a common activity or someone at random Performs an activity, choosing as follows:

17. Modeling the Effect of Social Interaction  Parameters Θ= (Φ,Ψ,α,β,γ,δ,κ)  State of the model  Φ is the ratio of user discussion page edits to total edits  β is the proportion of edits that create new articles  is the concatenation of all users ’ activity instances  is all users in proportion to the number of activities they ’ ve undertaken  E k (u) is the k most recent instances of activities for each user u  N(u) is u ’ s past interactions  are the k most recent users to have performed activity a  P(U = u) denote the probability of sampling a given object u from the sequence U

18. Modeling the Effect of Social Interaction  First, we consider the probability of an event in which user u chooses to interact with user v, which we denote C u,v  f(u,v|S) is the probability that u chooses v by first sampling an activity from u ’ s history and then sampling v from the users engaged in that activity, marginalizing over activities a

19. Modeling the Effect of Social Interaction  Second, we consider the probability of u doing an activity a, denoted D u,a  g(u,a|S) denotes the probability of u picking some user v from N(u) and then picking activity a from E k (v), marginalized over users v  h(a|S) is an indicator function that is 1 if and only if a is a new activity.

20. Modeling the Effect of Social Interaction  We used the model to simulate user behavior in Wikipedia Simulated Wikipedia resultActual Wikipedia result

21. Modeling the Effect of Social Interaction  The vast majority of activities are edits (94%) rather than interactions (6%)  In choosing what to edit users are most influenced by the overall distribution of editing in Wikipedia (50%), then by their own edit history (35%), and finally by the the edit history of their neighbors (8%).  7% of the time they choose to start a new article rather than editing anything that already exists.  Neighbors ’ past activities influence people ’ s future behavior, but that this effect is less strong than people ’ s intrinsic interests.

22. Predictive value  The more friends in a community, the higher a user ’ s probability of joining that community  We compare similarity and social ties in predicting behavior Social-network links: the probability that a node adopts a given behavior given that k of its neighbors have done so. Similarity : the probability that a node adopts a behavior given that the k most similar nodes have already done so.

23. Predictive value  In Wikipedia, social ties are more predictive  In LiveJournal, interest similarity is more predictive

24. Predictive value  Wikipedia is focused on the creation of topic-specific articles of interest  LiveJournal is more focused on social interaction, which might lead one to expect the opposite results

25. Conclusions  We studied the interplay between selection and social influence in online communities  Modeled the feedback between activities and interactions  Compared social ties, similarity as predictors of behavior