Feature Selection and Weighting Methods in Sentiment Analysis

Sentiment analysis is the task of identifying whether the opinion expressed in a document is positive or negative about a given topic. Unfortunately, many of the potential applications of sentiment analysis are currently infeasible due to the huge number of features found in standard corpora. In this paper we systematically evaluate a range of feature selectors and feature weights with both Naı̈ve Bayes and Support Vector Machine classifiers. This includes the introduction of two new feature selection methods and three new feature weighting methods. Our results show that it is possible to maintain a state-of-the art classification accuracy of 87.15% while using less than 36% of the features.