Feature Extraction and Opinion Mining in Online Product Reviews

Abstract

In this era of web applications, web shopping portals have become increasingly popular as they allow customers to buy products from home. These websites often request the customers to rate their products and write reviews, which helps the manufacturers to improve the quality of their products and other customers in choosing the right product or service. The rapid increase in the popularity of e-commerce has increased the number of customers in these type of web-shopping portals, leading to an enormous number of reviews for each product or service. Each of these reviews may describe the different features of the products. Hence, the customer has to go through a large number of reviews before s/he can arrive to a fully informed decision on whether to buy the product or not. In this paper, we describe a system, which automatically extracts the product features from the reviews and determines if they have been expressed in a positive or a negative way by the reviewers. The proposed algorithm works in two steps, viz feature extraction and polarity classification. We use association rule mining to identify the most characteristic features of a product. In the second step we develop a supervised machine learning algorithm based polarity classifier that determines the sentiment of the review sentences with respect to the prominent features. Our experiments on the benchmark reviews of five popular products show that our classifier is highly efficient and achieves an accuracy of 79.67%. We did not make use of any domain specific resources and tools, and thus our classifier is domain-independent, and can be used for the similar tasks in other domains.