CMPS 245, Winter 17 Project: Ideology-Backed Stance Classification

The determination of stance in two-sided discussions/debates in online debate forums is a new and interesting problem in opinion mining. We target the task of classifying stance in ideological debates on online debate forums, and this is a rather challenging problem due to the nature of debate setting and the language used. (Hasan and Ng, ) had introduced the notion of using ideology as an inter-post, extra-linguistic constraint which are implemented using Integer Linear Programming. In this work, we seek to improve (Sridhar et al., 2015) approach on jointlymodeling disagreement and stance in online debate forums by integrating ideology as a latent variable in the joint model. We show improvement of a few percentage points over the baseline set by (Sridhar et al., 2015). Additionally, we have also created a new ideology corpus of around 1.5 million tweets by 645 official members of the U.S. Congress.