A Hybrid Approach to Predict Ratings for Book Recommendation System Using Machine Learning Techniques

Abstract

A recommender system is a tool that suggests products or services to users based on their preferences and past behavior, enhancing user satisfaction and engagement. Accurate rating prediction is crucial as it directly impacts the system's ability to provide relevant and personalized recommendations, thereby improving the overall user experience. In this study, we introduce an innovative approach to recommendation systems by proposing an Weighted Hybrid Model that combines an Adaptive K-Nearest Neighbors (AKNN) algorithm and Singular Value Decomposition (SVD). The AKNN algorithm dynamically adjusts the number of neighbors based on user rating density, providing a tailored neighborhood size for each user. By incorporating a hybrid similarity measure that combines cosine similarity, Pearson correlation, and Variance Mean Difference (VMD), our AKNN algorithm effectively captures the multifaceted nature of user-item relationships. We further enhance our recommendation model by combining AKNN with SVD through optimized weighting, creating a Weighted Hybrid Model. This model balances the contributions of the AKNN and SVD components, leveraging the strengths of both approaches to minimize prediction errors. Our evaluation results demonstrate that the Weighted Hybrid Model outperforms several algorithms, including standalone KNN with Z-Score, Item-wise Variance-Mean based Recommender System (IVMRS), KNN with RJAC DUB, and Pearson Baseline with Weighted KNN. The Weighted Hybrid Model achieved the lowest Root Mean Squared Error (RMSE) of 1.54491 and Mean Absolute Error (MAE) of 1.17839, indicating superior predictive accuracy. © 2024 IEEE.