Fast Collaborative Filtering with a k-nearest neighbor graph

Traditional user-based/item-based Collaborative Filtering algorithms predict the preferences of all of the unseen items of a user. While this approach facilitates evaluations of the accuracy of various algorithms using the root mean square error, it consumes a considerable amount of time to recommend items for users. In this paper, we present a fast Collaborative Filtering algorithm using a k-nearest neighbor graph. Not only does this algorithm predict the preferences of only the k-nearest neighbor items, but it also shortens the execution time by calculating a k-nearest neighbor item graph in less time based on greedy filtering. The experimental results show that our approach outperforms traditional user-based/item-based Collaborative Filtering algorithms in terms of both the preprocessing time and the query processing time without sacrificing the level of accuracy.