A fast collaborative filtering algorithm for implicit binary data

Item-based and user-based collaborative filtering are two well-known algorithms for recommender system in e-commerce. Both the algorithms make use of similarity matrix whose elements represent the similarity of each item pairs or user pairs. A fast algorithm for item-based similarity matrix computation using cosine similarity metric was reviewed and applied for user-based one with some modification. The results show that the fast algorithm can blend well with other similarity metrics, and it can greatly improve the computational performance.