Language-model-based pro/con classification of political text

Given a controversial political topic, our aim is to classify documents debating the topic into pro or con. Our approach extracts topic related terms, pro/con related terms, and pairs of topic related and pro/con related terms and uses them as the basis for constructing a pro query and a con query. Following standard LM techniques, a document is classified as pro or con depending on which of the query likelihoods is higher for the document. Our experiments show that our approach is promising.