Aspect-Based Product Recommendation System by Sentiment Analysis of User Reviews

Abstract

Over the past decade, online shopping has become the norm. Everything that a person needs is available on online platforms such as Amazon and Flipkart. There is no dearth in the quantity and categories of products available on these online platforms. Given this influx of products in these online marketplaces and the rapid shift shown by people toward online shopping, it becomes imperative for an efficient product recommendation system to be in place in order to make product recommendations for a user as effective and seamless as possible. In this paper, a method for a recommendation system that is based on the aspects present in the user inputs and reviews is proposed, implemented, and evaluated. As opposed to just depending on user ratings and the sentiment scores and orientation (positive versus negative) of user reviews to recommend products, the proposed method performs aspect extraction of the given user reviews, calculates the sentiment associated with the aspects along with its positive and negative connotation and orientation and based on this, recommends relevant products to the user as per the requirements entered by the user in our system. The classifiers used for this purpose are k-NN, SVM, SentiWordNet, and NLTK VADER. Thus, based on the product requirements entered by the end user in our recommendation system, relevant products are recommended to the user.