Classifying Stances of Interaction Posts in Social Media Debate Sites.

Social media debate sites provide a rich collection of public opinions on various controversial issues. In an online debate, participants express their stances by replying directly to the main topic or indirectly to other participants. We observe that the majority of the posts in online debates are interaction posts. In this paper, we propose a new method for the task of stance classification of interaction posts in online debates. We mine the historical activities and textual content of posts to learn the relationships between participants in online debates. Then we build the interaction graph and develop a greedy algorithm to classify participants by stance. Empirical evaluation shows that our method performs better than the baseline methods.