User-sentiment topic model: refining user's topics with sentiment information

ABSTRACTIn large social networks, users feel free to share their feelings about anything they are interested in and many research works have focused on modeling users' interests on social network for product recommendations or personal services. Unfortunately, there are fewer works about finding why users like or dislike something. More specifically, there are many researches about sentiment analysis of users' opinion toward products or topics, but fewer are focused on why they hold this feeling and which aspects or factors of the product (topic) lead to users' different opinions about it. In this paper, we present a hierarchical generative model, called user-sentiment topic model (USTM), which captures users' topics with sentiment information. Our aim is to use USTM to refine users' topics with different sentiment trends including positive, negative and neutral, which can be further used in social network analysis to find influential users on topic level with sentiment information. The experiment results on three datasets show that our proposed USTM can capture user's interests with their sentiment well, making it useful for social network analysis.