New term weighting schemes with combination of multiple classifiers for sentiment analysis

Abstract

The rapid growth of social media on the Web, such as forum discussions, reviews, blogs, micro-blogs, social networks and Twitter has created huge volume of opinionated data in digital forms. Therefore, last decade showed growth of sentiment analysis task to be one of the most active research areas in natural language processing. In this work, the problem of classifying documents based on overall sentiment is investigated. The main goal of this work is to present comprehensive investigation of different proposed new term weighting schemes for sentiment classification. The proposed new term weighting schemes exploit the class space density based on the class distribution in the whole documents set as well as in the class documents set. The proposed approaches provide positive discrimination on frequent and infrequent terms. We have compared our new term weighting schemes with traditional and state of art term weighting schemes. Some of our proposed terms weighting schemes outperform the traditional and state of art term weighting schemes results.