Aspect-Based Sentiment Words and Their Polarities Using Chi-Square Test

Abstract

Most of the user-preferred products on e-commerce websites are accompanied by a massive number of product reviews and manually analyzing each review to understand the features and the user opinions associated with the products is an inconceivable task. A single domain of products can contain thousands of different products and an equally significant number of associated features, thereby making the polarity of the sentiment words in the product reviews vary widely according to the feature with which they are associated. The paper attempts to automatically calculate the feature-specific polarity of the sentiment words in a given domain using the Chi-square Test statistical measure. The results of the method are tested on two different domains. The experimental results show that the method delivers an accuracy of more than 75\% in both domains tested.  The method also helps in discovering strong domain-specific polar adjectives specific to a domain that  might be missing from the universal sentiment lexicon.