Comparison of Feature Selection Methods for Sentiment Analysis

Sentiment analysis is a sub-field of Natural Language Processing and involves automatically classifying input text according to the sentiment expressed in it. Sentiment analysis is similar to topical text classification but has a significant contextual difference that needs to be handled. Based on this observation we propose a new feature selection method called Document Frequency Difference to automatically identify the words which are more useful for classifying sentiment. We further compare it to three other feature selection methods and show that it can help improve sentiment classification performance.