Improving the performance of the lexical approach for sentiment analysis by introducing polarity calculation rules

Abstract:

Using of social networking data to study customer‟s attitudes toward products, services, or events has become more and more dominant in business strategic management research. Moreover, the reviews are not only useful for business to understand their customers and to improve their services, but also necessary for customers to seek for advice. In this context, sentiment analysis aims at extracting opinions and sentiments from natural language text using computational methods. It is the most common text classification tool that allows to classify the sentiment in a positive, negative or neutral categories. In this paper, we present a lexicon based approach to solve sentiment polarity categorization problem for online product reviews collected from Amazon.com. The proposed approach includes the following steps: Preprocessing, negation handling and polarity calculation. It relies on lexical dictionary of positive and negative polarity. In addition, we propose a rule based method to improve the polarity classification by considering contextual information and specific words lists such as modifier and intensifier. The results of experiments show that the proposed method provide good performance for sentiment classification. The value of F-measure is 86.62% for the mobile phone reviews dataset.