中文摘要：

针对矩阵分解推荐算法在潜在属性与已知属性之间不能建立对应关系的问题,提出了一种混合显式属性与隐式属性的矩阵分解算法。该算法使用显式属性的相关性对因子矩阵进行约束,能够抑制稀疏数据矩阵分解中过拟合的问题,提高推荐精度;由于因子矩阵中包含显式属性,混合因子矩阵分解算法可以实现对新用户和新产品推荐,部分解决冷启动问题;该算法实现了从评分数据到显式属性的映射,对推荐结果能够给出一定的解释。在Movie Lens数据集上的实验结果表明：在相同因子数目的情况下,混合因子矩阵分解算法的推荐精度均优于BPMF算法,并能够基于显式属性实现对新产品的推荐。

英文摘要：

A novel hybrid matrix factorization algorithm （HMF） is proposed to solve the problem that the correlation between latent factors and explicit attributes can not be established in traditional matrix factorization methods. The algorithm combines implicit and explicit attributes and uses correlations among explicit attributes to constrain factor matrixes, and to relieve the over fitting in sparse data matrix decomposition. Since factor matrixes include explicit attributes, HMF is used to solve the problem of cold start and to recommend new items. HMF realizes mapping from rating matrix to weights of explicit attributes and offers an interpretation for recommender items. Experiment on MovieLens clatasets shows that the accuracy of HMF is superior to that of BPMF for same number of factors, and HMF can be used to recommend new items based on explicit attributes.