Bootstrapping an Unsupervised Approach for Classifying Agreement and Disagreement

People tend to have various opinions about topics. In discussions, they can either agree or disagree with another person. The recognition of agreement and disagreement is a useful prerequisite for many applications. It could be used by political scientists to measure how controversial political issues are, or help a company to analyze how well people like their new products. In this work, we develop an approach for recognizing agreement and disagreement. However, this is a challenging task. While keyword-based approaches are only able to cover a limited set of phrases, machine learning approaches require a large amount of training data. We therefore combine advantages of both methods by using a bootstrapping approach. With our completely unsupervised technique, we achieve an accuracy of 72.85%. Besides, we investigate the limitations of a keyword based approach and a machine learning approach in addition to comparing various sets of features