Online and semi-online sentiment classification

Abstract

With the advent of social media and e-commerce sites, people are posting their unilateral, possibly subjective views on different products and services. Sentiment classification is the process of determining whether a given text is expressing positive or negative sentiment towards an entity (product or service) or its attributes. In this regard, we employed text mining involving steps like text preprocessing, feature extraction and selection and finally classification by machine learning algorithms to classify the customers' reviews on four mobile phone brands. The trio of TF-IDF, chi-square based feature selection and recurrent (Jordan/Elman)neural network classifier outperformed all other alternatives. The proposed combination yielded 19.13% higher accuracy compared to that of SVM, which is reported as the best classifier for sentiment classification in several studies. It also outperformed two semi-online classifiers proposed by us here. © 2015 IEEE.