SenticNet-Based Feature Weighting Scheme for Sentiment Classification

Abstract

Sentiment analysis aims to study the peoples' perceptions, sentiments and emotions in comments, blogs, review and so on. The performance of classifier can be improved by the use of efficient feature weighting schemes. The existing feature weighting schemes are based on frequency. They do not consider the meaning of the words for weighting. This leads to the degradation in the performance of machine learning algorithms. To address this issue, in the current study, SenticNet is used. SenticNet is a sentiment lexicon resource extracted from ConceptNet consisting of the polarity value in a range from − 1 to + 1. Classifiers like Naive Bayes and support vector machine are used to classify the reviews. The performance of SenticNet and existing feature weighting schemes are evaluated for sentiment classification. From the experiments conducted, it is found that SenticNet-based feature weighting scheme outperforms the existing schemes like binary, TF and TF-IDF.