Social Networks Influence and Propagation for User Preference Prediction

Nowadays, people have become increasingly dependent on social networks. By means of social networks, people can make friends, get information and make purchases, etc. Social networks have been widely applied in user behavior analysis, preference prediction and recommendation as people's decisions are influenced by their social relationships. However, static social relationship in a network alone is insufficient to model interpersonal influence and predict user preferences. In this paper, we propose to use user interaction records for modeling influence propagation and providing recommendation. Specifically, we propose a local user interaction network topology (LUINT) model to calculate the social influence between neighbors, which takes into account three types of user interactions: "at" action, comment, and re-tweet. Moreover, we design and adopt a shortest path with maximum propagation (SPWMP) algorithm to model the influence propagation within the network. To evaluate our approach, experiments on dataset KDD Cup 2012, Track 1 are conducted. The results indicate that the proposed model significantly outperforms the other benchmark methods in predicting preference of the users.