Enhanced Hybrid UBCF-IBCF Recommender Systems Using Pearson and Cosine Similarities for Improved Accuracy

The importance of recommender systems lies in its ability to offer tailored recommendations, therefore augmenting user happiness. This paper presents an improved novel hybrid recommender system that integrates User-Based Collaborative Filtering (UBCF) and Item-Based Collaborative Filtering (IBCF) by utilizing Pearson and Cosine similarity metrics. The goal is to enhance the precision of predictions and facilitate more effective recommendations. Using performance criteria such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Squared Error (MSE), we assess the proposed hybrid models on the MovieLens 100k and 1M datasets. The results of our comparison research indicate that hybrid models exhibit superior accuracy compared to individual UBCF and IBCF models. The findings indicate that the incorporation of several similarity measures can greatly improve the effectiveness of recommender systems. Our results show that hybrid models produce low MAE of 0.919, in 100k dataset, and 0.907, in 1M dataset, producing lower error rates compared to traditional models. This research presents a novel blend of Pearson and Cosine similarity metrics within a hybrid UBCF-IBCF architecture to attain high precision and low error rates. This solution solves the restrictions of existing collaborative filtering algorithms, such as data sparsity and the cold-start problem.