个性化服务中基于用户聚类的协同过滤推荐

Collaborative filtering recommendation based on user clustering in personalization service

协同过滤技术被成功地应用于个性化推荐系统中,但随着系统规模扩大,它的效能逐渐降低。针对此缺点,使用了基于用户聚类的协同过滤推荐,根据用户评分的相似性对用户聚类,在此基础上搜索目标用户的最近邻居,从而缩小用户的搜索范围。本文还提出将协同过滤推荐分为类内相似系数计算和产生推荐两个阶段,把相似系数的计算放在离线部分,减少在线推荐的计算量,提高实时响应速度。另对聚类算法初始聚类中心的选取也做了改进。

Collaborative filtefing is the most successful technology for building recommendation systems. Unfortunately, the efficiency of this method declines linearly with the number of users and items. A collaborative filtering recommendation algorithm based on user clustering was employed to solve this problem. Users were clustered based on users＇ ratings on items, then the nearest neighbors of target user can be found in the user clusters most similar to the target user. Based on the algorithm, this paper proposed that the collaborative filtering algorithm should be divided into two stages： to compute the similar coefficient and to produce recommendation. The first stage was done in the off-line phase and thus the computation in the on-line recommendation phase was reduced and the speed of on-line recommendation system was increased. And this paper also improved the initial center point＇s selection of K-Means clustering algorithm.