二分网络中基于谱聚类的协同推荐

Collaborative recommendation based on spectral clustering in bipartite network

提出一种基于谱聚类的协同推荐算法(SCBCF)。首先从用户——项目二分网络的单顶点投影中得到用户之间的相似矩阵,然后对该矩阵应用谱聚类算法,将用户聚成k类,并将得到的聚类结果用于数据平滑和邻居结点的选择,最后基于最近邻居集评分行为,对目标用户产生推荐。在Movie-Lens上的实验结果证明本文方法比传统的协同过滤算法能更好地应用于二分网络的协同推荐。

This paper proposes an approach based on spectral clustering for collaborative recommendation.Firstly,we conduct user-user similarity matnx from item-user bipartite network through one-mode projection. Then we apply spectral clustering method to cluster users into k clusters from the user-user similarity matrix for data smoothing and neighborhood seleetion.At last,we make recommendation for target user based on the rating behavior of nearest neighbors set.Experiment result on MovieLens shows that our proposed approach is better than traditional collaborative filtering algorithm.