Text-based feature selection using binary particle swarm optimization for sentiment analysis

The upsurge in social media data due to the proliferation of Web 2.0 applications has escalated scholarly studies within the sentiment analysis domain in recent times. Sentiment Analysis usually considered a text classification task in Natural Language Processing (NLP) classifies the views, attitudes, and feelings expressed by people concerning a particular organization or entity. This unstructured textual data can be pre-processed and represented as feature vectors which then serve as input to a machine learning algorithm for sentiment classification. In this process, feature selection which is a binary problem becomes an essential component of the SA exercise. We present a metaheuristic-based approach for optimal selection of features subset via the binary particle swarm optimization (BPSO) metaheuristic algorithm with the view to improve sentiment classification accuracy on the sentiment labelled sentences benchmark dataset. K-Nearest Neighbours, Naïve Bayes, and Support Vector Machine classifiers were employed as baseline classifiers to train the features. Before the sentiment classification process, the BPSO is utilized for selecting the optimal text features subset from the data. We train our sentiment labelled sentences benchmark dataset with SVM, NB, and k-NN using the selected optimal feature subset for sentiment classification. The results of the experiments conducted show impressive performance using our proposed approach for optimal text feature selection and sentiment classification compared to the baseline classifiers.