Let’s Tango – Finding the Right Couple for Feature-Opinion Association in Sentiment Analysis

Abstract

One approach in opinion mining is to perform sentiment classification at the sentence level. User’s view on a discovered product feature is predicted by the opinion words, e.g. adjectives, appeared in the same sentence. A number of previous works has been proposed and these approaches typically treat the feature and word relations identically. Blindly using sentiments of all opinion words to perform classification would lead to false results. In this paper, we investigate the relationship between features and opinion words using the corpus-based approach. We proposed a Feature-Opinion Association (FOA) algorithm to match these two in sentences to improve sentiment analysis results. We construct a feature-based sentiment lexicon using the proposed algorithm in the sentiment identification process. Extensive experiments based on a commercial product review site show that our method is quite effective in obtaining a more accurate result.