Integrated topic modeling and sentiment analysis: a review rating prediction approach for recommender systems

Abstract

Recommender systems (RSs) are running behind E-commerce websites to recommend items that are likelyto be bought by users. Most of the existing RSs are relying on mere star ratings while making recommendations.However, ratings alone cannot help RSs make accurate recommendations, as they cannot properly capture sentimentsexpressed towards various aspects of the items. The other rich and expressive source of information available that canhelp make accurate recommendations is user reviews. Because of their voluminous nature, reviews lead to the informationoverloading problem. Hence, drawing out the user opinion from reviews is a decisive job. Therefore, this paper aims tobuild a review rating prediction model that simultaneously captures the topics and sentiments present in the reviews whichare then used as features for the rating prediction. A new sentiment-enriched and topic-modeling-based review ratingprediction technique which can recognize modern review contents is proposed to facilitate this feature. Experimentalresults show that the proposed model best infers the rating from reviews by harnessing the vital information present inthem.