Feedback Effects between Similarity and Social Influence in Online Communities

Description Click to Edit

People rapidly become more similar shortly before their first communication and continue to become more similar for a long time afterward. In other words, social interaction is both an effect and a cause of selection, and theories and models that relate them will need to consider their interaction. Second, we find strong evidence that people become aware of others through shared, recent activity around artifacts. This parallels the relationship between social interaction and selection in the physical world: people are more likely to talk to others they encounter in the same church, school, or workplace.

Social influence and selection produce homogeneity in ways that have very different structural consequences for the network: social influence can produce network-wide uniformity, as a new behavior spreads across the links, while selection tends to drive the network toward smaller clusters of like-minded individuals, a process sometimes called balkanization. Moreover, because the two forces are based on different effects — interaction and similarity, respectively — the distinctions between them reflect analogous contrasts that arise in current computing applications
that mine social network data. In particular, applications such as viral marketing are rooted in the premise that a person’s social contacts provide a valuable predictor of their
future behavior, while recommender systems, which often build predictions based on the behavior and opinions of others who share similar behaviors and opinions.

People rapidly become more similar shortly before their first communication and continue to become more similar for a long time afterward. In other words, social interaction is both an effect and a cause of selection, and theories and models that relate them will need to consider their interaction. Second, we find strong evidence that people become aware of others through shared, recent activity around artifacts. This parallels the relationship between social interaction and selection in the physical world: people are more likely to talk to others they encounter in the same church, school, or workplace.

Social influence and selection produce homogeneity in ways that have very different structural consequences for the network: social influence can produce network-wide uniformity, as a new behavior spreads across the links, while selection tends to drive the network toward smaller clusters of like-minded individuals, a process sometimes called balkanization. Moreover, because the two forces are based on different effects — interaction and similarity, respectively — the distinctions between them reflect analogous contrasts that arise in current computing applications
that mine social network data. In particular, applications such as viral marketing are rooted in the premise that a person’s social contacts provide a valuable predictor of their
future behavior, while recommender systems, which often build predictions based on the behavior and opinions of others who share similar behaviors and opinions.

Original URL Click to Edit

http://www.cs.cornell.edu/home/kleinber/kdd08-sim.pdf

http://www.cs.cornell.edu/home/kleinber/kdd08-sim.pdf