Collaborative filtering recommender systems: Methods, strengths and weaknesses

Recommender systems are indispensable in e-business due to the extensive product range and the large number of niche articles. Collaborative filtering (CF) algorithms play a central role in generating personalised recommendations. There are numerous CF approaches in the literature that have specific advantages and limitations. The aim of this article is to provide a comprehensive overview of the characteristics of model- and memory-based CF methods through a systematic literature analysis of 62 scientific papers. The analysis shows that memory-based approaches are convincing due to their simple implementation and interpretability but exhibit scaling problems as well as susceptibility to data sparsity and the cold start problem. Model-based methods, on the other hand, offer greater scalability and robustness against data sparsity, but require a more complex implementation, higher computing power and complicate the interpretation of results.