A Sentiment and Topic Model with Timeslice, User and Hashtag for Posts on Social Media

Nowadays plenty of user-generated posts, e.g., tweets, are published on social media and the posts imply the public’s opinions towards various topics. Joint sentiment/topic models are widely applied in detecting sentiment-aware topics on the lengthy documents. However, the characteristics of posts, i.e., short texts, on social media pose new challenges: (1) context sparsity problem of posts makes traditional sentiment-topic models infeasible; (2) conventional sentiment-topic models are designed for flat documents without structure information, while publishing users, publishing timeslices and hashtags of posts provide rich structure information for modeling these posts. In this paper, we firstly devise a method to mine potential hashtags, based on explicit hashtags, to further enrich structure information for posts, then we propose a novel Sentiment Topic Model for Posts (STMP) which aggregates posts with the structure information, i.e., timeslices, users and hashtags, to alleviate the context sparsity problem. Experiments on Twitter7 show STMP outperforms previous models in sentiment-aware topic extraction.