Impact of clustering on quality of recommendation in cluster-based collaborative filtering: an empirical study

摘要

In memory nearest neighbour computation is a typical approach for collaborative filtering (CF) due to its high recommendation accuracy. However, this approach fails on scalability; which is the declined performance of the same due to the rapid increase in the number of users and items in archetypal merchandising applications. One of the popular techniques to attenuate scalability issue is cluster-based collaborative filtering (CBCF), which uses clustering approach to group most similar users/items from complete dataset. In this work we present a detailed analysis of the impact of clustering in CF approach. Specifically, we study how the extent of clustering impacts collaborative filtering systems in terms of quality of predictions, quality of recommendations, throughput and coverage. Based on the empirical results obtained from two datasets, Movielens100K and Jester; we conclude that with increasing number of clusters the quality of predictions, the quality of recommendations and the throughput are enhanced but the coverage provided by clustered subsystems declines.