A Hybrid Recommendation Framework Leveraging User Community Detection and Latent Feature Extraction for Improved Personalized Recommendations

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

A recommender system enhances personalization by delivering information that matches user preferences, thereby improving the overall user experience across different applications. In recommender systems, the user-item rating matrix represents user preferences for a collection of items. In this work, we build a community hybrid recommender system using cosine similarity ( \(\text {CHRS}_{c}\) ) and a community hybrid recommender system using Pearson similarity ( \(\text {CHRS}_{p}\) ). The proposed framework incorporates community detection by grouping nodes into communities according to their similarities. However, identifying small communities within large-scale datasets remains a significant challenge. The existing Louvain community detection method has certain limitations, as it may fail to correctly detect disconnected communities in large networks. To address this limitation, we propose the \(\text {CHRS}_{c}\) and \(\text {CHRS}_{p}\) approaches. The proposed strategy consists of the following steps: (1) developing a bipartite graph from the user-item rating matrix, (2) applying the Leiden community detection algorithm to generate communities from the bipartite graph and creating separate rating matrices for each community, (3) performing a convex combination of Matrix Factorization with cosine similarity or Matrix Factorization with Pearson similarity for each community rating matrix along with item properties, and (4) evaluating the performance using root mean square error (RMSE), mean absolute error (MAE), precision, recall, and F1-score by comparing predicted and actual rating matrices. The proposed CHRS method is evaluated on standard benchmark datasets, including MovieLens 100K, MovieLens 1M, and Anime Recommendations. Experimental analysis on the Anime Recommendation dataset shows nearly a 7% improvement in MAE compared with the Louvain based method. The results demonstrate that the suggested CHRS method effectively identifies communities with high internal similarity, even when community sizes are small, leading to more accurate recommendations.