Computational approaches to debate stance analysis

Recent years have seen a surge of interest in the automatic analysis of social media posts in the natural language processing (NLP) community. This dissertation focuses on analyzing a particular type of social media posts that have received relatively little attention: posts written in response to an online debate topic. In a debate post, a user typically expresses her stance (e.g., support, oppose) with respect to the topic under debate, and presents evidences and reasons to either support her stance or argue against the opposing stance. Automatic analysis of debate stance is of practical significance: for example, policy analysts may be interested in knowing not only the percentage of people supporting and opposing a public policy, but also the primary reasons why people take a particular stance. Nevertheless, such an analysis presents new challenges to NLP researchers, owing to the fact that online debaters use colorful and emotional language to express their points, which may involve sarcasm, insults, and questioning another debater's assumptions and evidence. We present new computational approaches to two key problems in the automatic analysis of debate stances, namely, determining the stance taken by its authors in a given debate post and identifying the reasons and arguments the author presents to support her stance in the post.