FANS: a framework for feature selection in sentiment classification using a modified Firefly algorithm

Sentiment classification is a prevalent task in text mining in which a text classifies into positive, negative, or neutral classes. Sentiment classification is an essential issue of decision-making for people, companies, etc. Feature selection is the most influential stage in sentiment classification. Due to the NP-hard nature of the problem and a huge of existing texts, the traditional feature selection techniques, such as statistical techniques, generate sub-optimal solutions. Swarm intelligence algorithms are extensively devoted to optimization problems. These algorithms produce features by increasing the classification performance and decreasing the computational complexity and feature set size. In this study, the authors proposed a framework using the modified multi-objective Firefly algorithm, namely FANS (Firefly Algorithm Naive Bayes Sentiment). The two targets are decreasing the naive Bayes error classifier and the k-nearest neighbor. A neural network is used as the final classifier. The three datasets on Movie review and Twitter domains are applied to evaluate the FANS. The FANS outperform its counterparts regarding precision, accuracy, and recall. The FANS yields 96.88% precision, 97.65% accuracy, and 96.54% recall.