STAR: rating of reviewS by exploiting variation in emoTions using trAnsfer leaRning framework

Abstract

In this digital era, it is a common practice to check reviews about a service or product before it buying. A five star rating scale system provides much easier interface to the consumers for checking the reviews about the corresponding service or product, instead of just classifying the reviews as good, neutral and bad. Moreover, it is very common for a single review which can praise the product and criticize it as well. Even if two reviews over all, show the same sentiment. However, the order in which they praise or criticize a product, can make their star rating quite different. We have considered such observations to deploy our proposed STAR model, which addresses the above concerns by involving the variation of sentiment in reviews, to greatly affect the star rating performance. This work highlights a two Phases based novel approach using transfer learning framework to analyze the reviews by exploiting the variation in human being emotions. The experimental analysis shows that the STAR model outperforms the state-of-the-art models.