Analysis of a Book Recommendation Model Using Hybrid Filtering for User Cold-Start Scenarios

Rapid technological advancements have accelerated the growth of digital data, creating information overload that challenges users' ability to identify relevant content. Recommender systems have emerged as a key solution, yet they still struggle with fundamental issues such as data sparsity and the cold-start problem. The cold-start scenario, where new users or items have little to no interaction history, remains particularly difficult to address, as it limits the system's ability to infer preferences. Hybrid filtering, which combines collaborative filtering with content-based filtering, offers a promising approach to mitigate these limitations by leveraging both interaction data and metadata. However, while widely studied in warm-start conditions, its effectiveness in user cold-start scenarios remains underexplored. Here, we present a performance analysis of a hybrid filtering model for book recommendation under user cold-start conditions. Using the BookCrossing dataset enriched with metadata from the OpenLibrary API, the model was trained with the Weighted Approximate-Rank Pairwise (WARP) loss function and tuned for optimal feature selection, embedding dimensions, and learning rate. The proposed hybrid approach achieved an AUC of 80.3969%, precision @ 10 of 22.61%, hit rate @ 10 of 84.8485%, and coverage of 57.6144%, outperforming baseline methods. These results demonstrate that hybrid filtering can effectively address the user cold-start problem by providing accurate and diverse recommendations even in the absence of extensive user history.