协同过滤推荐中基于用户分类的邻居选择方法

Approach of neighbor selection based on user classification in collaborative filtering recommendation

为了提高推荐系统的推荐结果质量,找到目标用户恰当的邻居是协同过滤算法中非常关键的一个环节。网络中的用户可以分为专家型用户、可信用户与兴趣相似用户三个维度,由于不同类型的邻居对用户的影响及用户对不同邻居的依赖倾向的不同,因此利用岭回归分析估计用户对于这三类用户的主观倾向,即邻居选择权重,由此获得目标用户邻居集合,进而产生推荐,通过利用标准F1方法与传统推荐方法对比实验分析表明,推荐结果的质量显著提高;同时利用K-means方法对用户作聚类分析及类别之间的方差齐性分析,并与行为研究结果相对比,验证了推荐结果的可信性。

In order to improve the quality of recommendation results,selecting proper neighbors would be the important link in collaborative filtering.The user might be divided into three types,which was expertise,trustworthy and similarity,since there were differences between neighbors,the important of them could be differentiated from target users.As the target user,the importance and weight of these three types of neighbors would be analyzed by the method of ridge regression.As a result,the proper neighbors might be found for the target user.According to comparative experiments based on F1 method,it shows that the accuracy has been improved significantly.Meanwhile,through the K-means cluster analysis and LSD（least-significant difference）,the result coincides with behavior research＇s.