A Review of Collaborative Filtering Techniques in Recommender Systems

Collaborative filtering is one of the oldest and most widely used techniques in recommender systems. Instead of relying on item content, it leverages user behavior, such as ratings, clicks, or interaction histories, to predict user preferences. This paper reviews the main categories of collaborative filtering, including memory-based and model-based approaches, and presents the mathematical formulations underlying their prediction mechanisms. Key developments in the literature, such as user-based and item-based similarity models and matrix factorization techniques popularized during the Netflix Prize, are discussed. The paper further analyzes practical challenges including data sparsity, cold start, scalability, and popularity bias. Finally, potential improvements such as hybrid models and neural approaches are examined while highlighting the continued relevance of simpler methods in real-world systems.