Personalized microtopic recommendation with rich information

Abstract

Sina Weibo allows users to create tags enclosed in a pair of # which are called microtopics. Each microtopic has a designate page, and can be directly visited and commented on. Microtopic recommendation can facilitate users to efficiently acquire information by summarizing trending online topics and feeding comments with high quality. However, it is non-trivial to recommend microtopics to the users of Sina Weibo to satisfy their information needs. In this paper, we focus on personalized microtopic recommendation. Collaborative filtering based methods only utilize the user adoption matrix, while content based methods only use textual information. However, both of them can not achieve satisfactory performance in real scenarios. Moreover, auxiliary information on social media provides great potential to improve the recommendation performance. Therefore, we propose a novel hierarchical Bayesian model integrating user adoption behaviors, user item content information, and rich contextual information into the same principled model. We experiment with different kinds of textual and contextual information from both user and microtopic sides on a real dataset. Experimental results show that our model significantly outperforms a few baseline methods.