使用二分图网络提高协同推荐的准确性

Using Bipartite Network for Enhancement of Collaborative Filtering

协同过滤是推荐系统中广泛使用的最成功的推荐技术,但却面临着严峻的稀疏性问题。评分数据稀疏性使得最近邻搜寻不够准确,导致推荐质量较差。使用二分图网络缓解协同过滤推荐系统中的稀疏性问题,即将用户和项目抽象为二分图网络中的节点,重新分配项目资源并计算项目间资源贴近度,据此填充用户未评分项目,将稀疏评分矩阵转化为完全矩阵。采用近邻传播聚类对评分矩阵进行聚类,提高算法的可扩展性。最后提出了两种不同的在线推荐策略:(1)通过加权目标用户所在类的邻居用户评分产生推荐(BNAPC1);(2)通过各个类的总体偏好产生推荐(BNAPC2)。在MovieLens和Netflix数据集上进行了实验,结果表明BNAPC1的预测精度优于BNAPC2,且与其他几种常用的推荐算法相比仍具有一定优势。

Collaborative filtering is one of the most successful and widely used techniques among recommender systems.However,it suffers from serious problem in sparsity.Sparsity in ratings makes the formation of neighborhood inaccurate,thereby resulting in poor recommendations.In this paper,bipartite network was used to alleviate the sparsity problem in collaborative filtering.Users and items are mapped to nodes in bipartite network,and resources on items are redistributed.Resource approach degree between items is computed,and the original rating matrix is converted to complete matrix based on the resource approach degree.Then affinity propagation clustering was applied to cluster the rating matrix to improve the scalability of our approach.Finally,two different recommendation methods were presented.One is generating recommendations according to neighbors in the cluster which active user belongs to（BNAPC1）,and the other is generating recommendations according to clusters’ preferences（BNAPC2）.Experiments on MovieLens and Netflix datasets show that BNAPC1 is more accurate than BNAPC2,and is also superior to existing alternatives.