Early Identification of Personalized Trending Topics in Microblogging

Abstract

Social media has become a primary platform for the spread of information. Trending topics, which are breaking news and immediately popular stories, have become an attractive data source facilitating the spread of emerging issues. Motivated by the diverse trending topics covering from sports to politics, it is essential to help users find personalized trending topics. Since a topic in social media may start trending and get obsoleted quickly, the personalization would be more valuable to a user if the trending topic can be recommended before it is outdated. In order to identify personalized trending topics at an early stage, we propose to identify and exploit the auxiliary information. In particular, through collectively modeling content of similar users with social network information, we identify additional past contents that can enrich the training data of trending topics and users. The key insight is that though most posts of a user may be irrelevant, a few key posts can be signals revealing interests towards a particular topic. Experiments on real-world data demonstrate that our proposed approach effectively personalizes trending topics when they just start trending.