基于矩阵分解与用户社会关系的协同过滤推荐算法

Collaborative Filtering Recommendation Algorithm Based on Matrix Factorization and User Social Relationships

矩阵分解技术虽然已成功应用到协同过滤推荐系统中,但是基于矩阵分解的传统协同过滤方法仍然存在着数据稀疏性、冷启动等问题。为了进一步提高系统推荐的准确性,提出将用户间的社交关系融合到矩阵分解的协同过滤推荐系统中,该方法以奇异值矩阵分解推荐模型为核心,对该模型添加用户偏置和项目偏置,同时将用户在社交网络中的朋友关系添加到矩阵分解模型中,然后采用一种随机梯度下降法对该矩阵进行分解,得到用户潜在特征和物品潜在特征。最后通过实验结果验证表明,所提出的算法具有较好的预测效果,其性能明显优于现有的相关算法。

Although matrix factorization（MF） based method has been successfully applied to collaborative filtering（CF）. Recommender System（RS）,it still suffers from data sparsity, cold start and other issues. In order to further improve the accuracy of the CF, integrates the social relationship among users into MF based CF. Based on singular value decomposition（SVD）, the proposed method merged social friendships, the users rating bias and items rating bias into MF model. Designs a Stochastic Gradient Descent（SGD） algorithm to obtain potential user feature matrix and item feature matrix. Experimental results show that the proposed algorithm has good prediction accuracy,which is better than the existing algorithms.