$R^3N^2$: a novel approach for review based custom star rating using recurrent neural network

Abstract

Online shopping is growing in popularity as internet services are used more frequently. Many consumers consider online shopping due to the comfort, ease of access, and variety of products offered. However, this wide variety of products makes it hard for consumers to select one product from many available options. It is infeasible for any customer to go through many customer reviews given to the products. Therefore, the customers’ star rating becomes a viable solution for the star rating assigned to the products. However, the star rating-based system takes the mean of star ratings given by all the consumers. The mean value represents the masses’ view but fails to represent personal views. In this work, a novel star ratings-based system is proposed for the products to reflect the consumer’s individual needs and expectations by extracting customer reviews. A recurrent neural network is used as a sequential prediction model to hold aspect words and identify them after tokenization of a detailed review. The proposed $R^3N^2$ model enhances 0.9%, 2.5%, 4.9%, and 6.7% accuracy on digital products, electronics products, Kaggle movie reviews, and Apps datasets than the existing state-of-the-art (SOTA) models, respectively. Therefore, the proposed model outperforms SOTA models and may be used for real-time applications as it takes 2.56 ms per review.