Unifying Local and Global Agreement and Disagreement Classification in Online Debates.

Online debate forums provide a powerful communication platform for individual users to share information, exchange ideas and express opinions on a variety of topics. Understanding people's opinions in such forums is an important task as its results can be used in many ways. It is, however, a challenging task because of the informal language use and the dynamic nature of online conversations. In this paper, we propose a new method for identifying participants' agreement or disagreement on an issue by exploiting information contained in each of the posts. Our proposed method first regards each post in its local context, then aggregates posts to estimate a participant's overall position. We have explored the use of sentiment, emotional and durational features to improve the accuracy of automatic agreement and disagreement classification. Our experimental results have shown that aggregating local positions over posts yields better performance than non-aggregation baselines when identifying users' global positions on an issue.