<![CDATA[The rapid growth of digital information necessitates robust recommender systems to filter relevant content effectively. This study addresses the challenge of generating personalized book recommendations using a highly sparse dataset (>99% sparsity) derived from the Recommender System 2023 Challenge. The primary objective is to predict user interest based on implicit feedback and optimize the ranking quality of Top-N recommendations. We conduct a comparative analysis between a memory-based baseline, Item-Based Collaborative Filtering (IBCF), and a model-based approach, Alternating Least Squares (ALS). The methodology incorporates k-core decomposition for data preprocessing to mitigate sparsity issues, followed by rigorous hyperparameter optimization via grid search to tune latent factors, regularization parameters, and confidence weights. Experimental results demonstrate that the baseline IBCF model struggles significantly with data sparsity, yielding a Mean Average Precision at 10 (MAP@10) of only 0.0018. In contrast, the hyperparametertuned ALS model achieves a MAP@10 of 0.0348, representing a 19.3- fold improvement over the baseline and a 3.3-fold improvementover the default ALS configuration. These findings confirm thatmatrix factorization techniques, when systematically optimized,significantly outperform memory-based methods in handling sparseimplicit feedback, providing a scalable and accurate solution forranking tasks.]]>
Copyrights © 2026
