Improving performance of recommendation systems using sentiment patterns of user

Nowadays, shopping malls use tactics that affect customers psychologically. They arrange items based on human psychology. Likewise, the E-commerce industry aims to increase profits by recommending the most suitable items to customers. This paper contributes a novel approach to sentiment pattern recognition that identifies the change in user interests over time in Recommendation System (RS). The proposed work focuses on improving the performance of RS using historical sentiment patterns of user. This paper proposes a new User Sentiment Pattern-based Recommendation System (USPRS) that uses a Multi-Layer Perceptron (MLP) in collaborative filtering to identify a user's sentiment pattern from their selection history. By analysing a user's sequential history of item interests, the USPRS generates sentiment patterns to provide relevant recommendations using an MLP approach, which is useful for sparse datasets. The MLP model is developed to solve a regressive prediction problem. The proposed method is evaluated on natural and public datasets such as MovieLens, YOOCHOOSE, CiaoDVD and Amazon. The experimental results demonstrate that the USPRS model achieves accurate and relevant recommendations compared to state-of-the-art methods. The proposed model improves RMSE by 3% and precision by 2.5% over existing methods. The proposed model shows RMSE of 0.7912 and 1.015 on MovieLens and CiaoDVD datasets respectively, and precision of 0.6381 and 0.161 on YOOCHOOSE and Amazon datasets respectively. © 2023, The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management.