Sentiment Analysis and Feature-Based Mining of Customer Product Reviews

Abstract

Online reviews have the potential to provide an insight to the buyers about the product like its quality, performance and recommendations; painting a clear picture of the product in front of the future buyers. Sentiment Analysis is a computational study to extract subjective information from the text. In this research, data analysis of a large set of online reviews for mobile phones is conducted. Variegated techniques have been used to perform classification of the text, namely, Lexicon-based approach using sentiment dictionary, Supervised Learning using Naïve Bayes, Support Vector Machine (SVM) and Decision Tree classifiers, Deep learning and Feature-based extraction of reviews using association rule mining. The text is not only classified into positive and negative sentiments but also represents eight different emotions using lexicon-based approach. This delineated classification of reviews is helpful to evaluate the product holistically, hence enabling better-decision making for consumers. Supervised learning is performed using binary and multi-class classification. The performance of SVM is the best in both with higher accuracy in multi-label data. To improve the efficiency of classification, deep learning has been used which classifies the data into two classes. It extracts subjective meaning of the text along with the negation effect. Two different techniques, Vocabularybased Vectorization and Feature Hashing have been employed. Vocabulary-based Vectorization outperforms Feature Hashing and there is a considerable rise in accuracy as compared to supervised learning. Feature-based extraction is employed to determine the performance of various features of mobile phones which aids the customers to make well-informed decisions and also highlights various flaws on which a product designer can work upon to improvise the product.