一种聚类与kNN结合的协同过滤算法

Collaborative Filtering Algorithm Based on Clustering and kNN

随着电子商务的发展,推荐系统被广泛用于挖掘用户行为数据中的商业价值.基于kNN的协同过滤是经典的推荐算法,但存在两个主要问题:时间复杂度高以及使用单个距离度量导致预测精度低.本文提出了一种聚类与kNN相结合的协同过滤算法(C-kNN).在预处理阶段,使用M-distance将商品划分成多个簇.在评级预测阶段,只有簇内的项目作为距离计算和预测的候选邻居.在四个真实数据集上的实验结果表明,C-kNN比经典kNN在MAE和RMSE上均有可观提升.

With the development of e-commerce,recommendation systems are widely used to mine business value from user behavior data. Collaborative filtering based on kNN is the classical algorithm of the recommendation system,but there are two main problems:high time complexity,and the use of a single distance metric leads to low prediction accuracy. This paper proposes a clustering and kNN combined collaborative filtering algorithm( C-kNN). In the preprocessing phase,the items are divided into multiple clusters using the M-distance. In the rating prediction stage,only intra-cluster items are used as candidate neighbors for distance calculation and prediction. The experimental results on four common datasets show that C-kNN obtains a considerable improvement in both MAE and RM SE than the classic kNN.