Sentiment based multi-index integrated scoring method to improve the accuracy of recommender system

To the best of our knowledge, few studies have focused on the inconsistency between user ratings and reviews as well as natural noise management in recommender systems (RSs). To address these issues, this study introduces a sentiment based multi-index integrated scoring method to provide a reliable information input that reflects comprehensive user preferences for recommendation algorithms and facilitate improved performance. Initially, Bing Liu's lexicon is expanded using a semi-supervised learning technique to obtain additional sentiment words and calculate the sentiment scores of reviews; then a normalized sentiment score method based on sigmoid function that considers the emotional tendencies of different users in reviews is designed to convert the scores into values corresponding to the rating scale of RS. Subsequently, a degree classification criteria approach is adopted to assign users and items to more fine-grained classes Further, a natural noise detection method is exploited to identify and correct noise ratings according to classification conditions. To effectively integrate normalized review and denoised rating information, two factors, user consistency and review feedback, are considered to obtain the importance of reviews and ratings; then, a weighted average method is used to generate a set of comprehensive ratings. The experimental results on two benchmark datasets indicate that the superiority of memory-based or model-based collaborative filtering methods (CFs) using comprehensive ratings over their respective methods using original ratings is determined by various accuracy metrics, which demonstrates that our scheme can enhance the reliability and accuracy of user information. Thus, the proposed scheme provides new insights for improving the accuracy of RSs from the perspective of multiple information sources. Additionally, this method exhibits good generalizability and practicality.