Performance Comparison of Recommendation Models: IBCF and ALS Based on Implicit Feedback

Abstract

The rapid growth of digital platforms has led to an overwhelming amount of information, making it difficult for users to discover relevant books among the vast selection available. Recommendation systems, particularly collaborative filtering techniques, help address this issue by personalizing recommendations based on user interactions. This study compares the performance of three recommendation models: Item-Based Collaborative Filtering (IBCF) as a baseline, Alternating Least Squares (ALS), and ALS with hyperparameter tuning (ALS-Tuning) in the context of implicit feedback data. Using two evaluation metrics, Mean Average Precision at 10 (MAP@10) and Root Mean Squared Error (RMSE), the results demonstrate that ALS-Tuning outperforms both IBCF and the standard ALS model in both metrics. The ALS-Tuning model shows the highest improvement in ranking relevance (MAP@10) and prediction accuracy (RMSE), with hyperparameter tuning playing a significant role in enhancing the model's ability to uncover latent user preferences. While IBCF struggles with sparse data, ALS, especially with tuning, proves to be more effective in capturing latent patterns in implicit feedback data. This research provides empirical evidence supporting the use of ALS-Tuning as a robust approach for recommendation systems with sparse and implicit data, offering valuable insights for future development in personalized recommendation systems.