关键词: 协同过滤, 聚类, 算法可扩展性, 数据稀疏性, 平均绝对偏差

Abstract:

Collaborative filtering is one of the main technologies for the ecommerce recommendation systems. However, the lack of algorithm scalability and the sparsity of rating data challenge the gradual increase of users and items. A hybrid recommendation algorithm based on clustering and collaborative filtering is  employed to solve these problems. Firstly, the clustering algorithm is utilized to cluster items into several classes. The operations for one user following the clustering algorithm are limited within the interested classes of the user.This strategy improves the scalability of the recommendation algorithm and reduces the computation time. Secondly,an itembased algorithm is employed to compute the predictive values and insert high values into the original matrix in order to relieve the sparsity of the rating data. Finally,a userbased algorithm is used to attain the final predictive value,and then the recommendations are generated.The experimental results indicate that this algorithm can efficiently resolve these problems, and can improve the recommendation quality.

Key words: collaborative filtering;clustering;algorithm scalability;sparsity of data;MAE