A Study of Collocations in Sentiment Analysis

Abstract

Collocations are not merely frequently appearing word combinations (n-grams). Words in collocations have some kind of strong association among them. Collocations play an important role in various natural language processing (NLP) applications. Sentiment analysis is one of the growing areas of research in NLP because of its utilization in various business strategies. The present paper investigates collocations in positive and negative sentiments and their usefulness in sentiment analysis. We considered Amazon Products Review dataset for the purpose and analyzed positive and negative reviews separately. Different statistical techniques; Pointwise Mutual information (PMI), Chi Square test (Chi2), t-test, and likelihood ratio (LH) have been used to extract collocations from these texts and the common collocations have been extracted and analyzed. We found that collocation may be a potential feature for sentiment analysis.