A Survey of Collaborative Filtering Techniques

This paper surveys collaborative filtering (CF), one of the most successful approaches for building recommender systems, which uses the known preferences of a group of users to make recommendations or predict the unknown preferences of other users. It begins by laying out the core CF tasks and their main challenges—including data sparsity, scalability, synonymy, the gray sheep problem, shilling attacks, and privacy protection—and discusses possible solutions to each.

The survey then presents three main categories of CF techniques—memory-based, model-based, and hybrid algorithms that combine CF with other recommendation methods—offering representative algorithms for each category and analyzing both their predictive performance and their ability to address the identified challenges. Spanning basic techniques through the state of the art, it is intended to serve as a comprehensive roadmap for research and practice in the field.