Hybrid Feature Vector Space based Ensemble Machine Learning Approach for Sentiment Analysis on Amazon Product Reviews

Abstract

In recent era, people are getting more attracted to micro blogs and social media to share their daily activities and express feelings and opinions. Machine learning based sentiment analysis becomes immensely popular to judge the feelings about a particular content on how positive or negative their feelings and opinions are before taking important decisions. In this paper, we propose an effective and combined machine learning approach with an enhanced hybrid feature vector space of latent concepts and external information features. The latent concepts are prepared by a supervised machine learning approach, and the external information features, estimating the quality of the information shared in the documents, are classified by the unsupervised rule-based learning approach. A Random Forest ensemble method has been utilized to build a classifier model, and some standard performance measures such as accuracy, precision, recall, f1-score and Cohen’s Kappa value have been taken into account to analyze the performance. The novelty of this paper lies in the hybridization of feature vector space of latent concepts and external information features along with the Random Forest ensemble classifier. Based on the analyses, the proposed approach outperforms its counterparts as well as provides better outcomes against other solo latent concept-oriented approaches.