Microblog User Interest Modeling Based on Feature Propagation

This paper focuses on the problem of building the user interest graph on microblog. While previous efforts in interest modeling on microblog have focused on building a "bag-of-words" profile only based on his own content or his followees' for each user. They ignored the fact that users always released posts about the fields which they are expert in, but not really interested in. By introducing the side information, a novel method is proposed to settle this problem by analyzing the posts they receive and update. The interest features are extracted from the user's followee by Latent Dirichlet Allocation (LDA), and then they are propagated to the user with a weighting factor. For every user, our method gives probabilities of different fields they may be interested in and it can be used for calculating the probability of a new topic this user will be interested in. We exploit the collected experimental data provided by Sina Weibo API to validate our method.