Scalable Discovery of Contradicting Opinions in Weblogs

Weblogs are a popular means of information communication, where people discuss a variety of topics, and often times also express their opinions on these topics. In this work, we address the problem of analyzing the evolution of community opinions across time, as these are represented in the weblogs. In particular, we are interested in identifying topics and time windows, for which contradictory opinions have been expressed. We describe an approach for solving the above problem, which consists of the following steps. We first introduce a technique for topic and opinion extraction that operates at the sentence level. Then, we propose a novel measure for contradictions that can effectively aggregate the relevant information from the weblog posts. We discuss its properties, and show how it can be used to detect two different types of contradictions, namely, simultaneous contradictions, and change of sentiment. Finally, we describe an efficient data structure for answering queries related to contradiction detection, and show that it has the additional property of being incrementally maintainable. A detailed experimental evaluation of our approach with synthetic and real datasets demonstrates the applicability and efficiency of our techniques.