Description

• Sentiment analysis aims to determine the attitude of a speaker or a writer with respect to some topic or the overall contextual polarity of a document. • We can detect sentiment and topic simultaneously from text. • JST model achieved either better or comparable performance compared to existing semi-supervised approaches. • We focus on document-level sentiment classification for general domains in conjunction with topic detection and topic sentiment analysis, based on the proposed weakly-supervised joint sentiment-topic (JST) model. • This model extends the state-of-the art topic model latent Dirichlet allocation (LDA), by constructing an additional sentiment layer, assuming that topics are generated dependent on sentiment distributions and words are generated conditioned on the sentiment-topic pairs. • JST is weakly-supervised, where the only supervision comes from a domain independent sentiment lexicon. • JST can detect sentiment and topics simultaneously. • We suggest that the weakly-supervised nature of the JST model makes it highly portable to other domains for the sentiment classification task.