Context-dependent sentiment classification using antonym pairs and double expansion

Abstract

Sentiment classification at word-level plays a fundamental role in many sentiment-related tasks. Context-free sentiment classification circumvents the essential factors in language use and is error prone due to oversimplification in its assumption. Context-dependent sentiment classification poses new challenges to researchers. We propose a novel approach to automatically determine the contextual polarity with the help of antonym pairs. First, neighboring nouns are extracted as context information within a predefined distance of sentiment words. Secondly, the polar posterior probabilities of sentiment words are derived based on Bayes' theorem. Finally, the polarity of one sentiment word with the context of one neighboring noun is determined by Bidirectional Rule and Unidirectional Rule. In addition, we define a new similarity measure, which combines semantic distance with edit distance, for double expansion, i.e., Context Expansion and Target Expansion. The experimental results on two real-world data sets validate the effectiveness of our approach. © 2014 Springer International Publishing Switzerland.