A new detailed rating strategy for enhancing online products recommendation using machine learning and sentiment analysis

Abstract

Abstract With the increased use of e-commerce platforms such as Amazon Marketplace, reliable and efficient recommendation of online products is required. High rated and excellent reviewed products by previous customers are considered highly recommended products for new customers. Products’ rating using overall numerical rates is the easiest and most commonly used strategy by most of e-commerce platforms. However, is not considered an accurate and representative rating method. Also, many online shopping platforms provide products’ text reviews of the previous customers, but it is difficult for new customers to read and analyze text reviews to take a decision of buying or not buying products based on their specific preferences and priorities. Sentiment analysis can be automatically applied on these text reviews, but it also generates an overall satisfaction of products which is not accurate for customers with different preferences and priorities. So, this paper proposed a new detailed products’ rating based on different satisfaction factors such as quality, price, speed, and usability for significantly and accurately helping customers to buy products based on their specific preferences and priorities. Also, sentiment analysis can be applied according to the proposed detailed rating. Real Amazon product reviews datasets were used. Proposed detailed rating is assigned to datasets based on analyzing text reviews. Different baseline models were used for ensuring model generalization and robustness of the proposed detailed rating strategy. The experimental results ensure the high performance of recommendation based on the proposed detailed rating methodology. The proposed recommendation methodology achieved accuracy up to 97.74% for some satisfaction factors, and average accuracy of 93.26%. Also, the Mean Square Error, the Root Mean Square Error, and the Mean Absolute Error measures’ values are closer to zero, which ensure the high prediction accuracy of the proposed recommendation system.