Detecting Agreement and Disagreement in Political Debates

摘要

In this paper, the task of agreement/disagreement detection in political debates is studied. The main goal of this study is to detect agreement/disagreement between two individuals on a topic based on their conversations. This is a challenging task due to the lack of annotated corpora in this field. A self-labeling method is introduced for data collection and generating the training data. A new approach based on text classification is proposed for this task. The experimental results on Canadian Parliamentary debates and the United State 1960 Presidential Campaign datasets have proven the efficiency of the developed methodology and outperforms the baseline methodologies. In addition, the validity of the proposed self-labeling method is evaluated, and its efficiency is confirmed.