Determining political interests of issue-motivated groups on social media: joint topic models for issues, sentiment and stance

Stance detection is an emerging research problem in opinion mining where the aim is to automatically determine from the text whether the author is for, against or neutral towards a proposition or target. In this paper, we propose a novel weakly supervised probabilistic topic model, Joint Issue-Sentiment-Stance Topic (JISST) model, for stance detection from political opinion in social media. The model automatically identifies the target issue and stance toward the target issue simultaneously from the text. Unlike other machine learning approaches to stance classification which require labelled data for training classifiers, JISST requires only a small number of seed words for each issue and stance and a sentiment lexicon. The model is evaluated on two datasets in the political domain: a Facebook dataset which contains posts of politically motivated Facebook groups in Australia and a Twitter dataset which was published for the SemEval 2016 competition. Experimental results demonstrate that JISST outperforms both weakly supervised and supervised baselines for stance and issue classification.