Optimizing Collaborative Filtering by Interpolating the Individual and Group Behaviors

Abstract

Collaborative filtering has been very successful in both research and E-commence applications. One of the most popular collaborative filtering algorithms is the k-Nearest Neighbor (KNN) method, which finds k nearest neighbors for a given user to predict his interests. Previous research on KNN algorithm usually suffers from the data sparseness problem, because the quantity of items users voted is really small. The problem is more severe in web-based applications. Cluster-based collaborative filtering has been proposed to solve the sparseness problem by averaging the opinions of the similar users. However, it does not bring consistent improvement on the performance of collaborative filtering since it produces less-personal prediction. In this paper, we propose a clustering-based KNN method, which combines the iterative clustering algorithm and the KNN to improve the performance of collaborative filtering. Using the iterative clustering approach, the sparseness problem could be solved by fully exploiting the voting information first. Then, as a smoothing method to the KNN method, cluster-based KNN is used to optimize the performance of collaborative filtering. The experimental results show that our proposed cluster-based KNN method can perform consistently better than the traditional KNN method and clustering-based method in large-scale data sets.