Building Phrase Polarity Lexicons for Sentiment Analysis

Abstract

Many approaches to sentiment analysis benefit from polarity lexicons. Most polarity lexicons include a list of polar (positive/negative) words, and sentiment analysis systems attempt to capture the occurrence of those words in text using polarity lexicons. Although there exist some polarity lexicons in many natural languages, most languages suffer from the lack of phrase polarity lexicons. Phrases play an important role in sentiment analysis because the polarity of a phrase cannot always be estimated based on the polarity of its parts. In this work, a hybrid approach is proposed for building phrase polarity lexicons which is experimented on Turkish as a low-resource language. The obtained classification accuracies in extracting and classifying phrases as positive, negative, or neutral, approve the effectiveness of the proposed methodology.