Sentiment Analysis of Microblog text based on joint sentiment-topic model

Sentiment Analysis of Microblog text is a challenging work. Microblog text is different with general user-generated text, because it often contains some special symbols such as “@ # //” and a lot of emotional symbols. Previous joint sentiment-topic models did not consider these features of Microblog texts and then could not well model them. In this paper, considering the structural features and content features of Microblog text, we present a semi-supervised joint sentiment-topic model (MB-PL-ASUM). This new model uses semi-structured information and emotional symbol information to classify sentiments of Microblog text without labeling them. The experiments of sentiment classification on real Sina Microblog texts show that MB-PL-ASUM outperforms word matching, JTS and ASUM model.