信任传递的矩阵分解推荐算法

Matrix factorization recommender algorithm using trust propagation

针对协同过滤推荐系统中数据稀疏性导致推荐准确性低下问题,提出信任传递的矩阵分解推荐算法.该算法利用用户社交网络的直接信任关系,基于信任传递思想,预测用户在社交网络中的间接信任关系,以解决社交网络信任关系的稀疏性问题.该算法使用填充后的社交网络信任数据,预测填充用户评分数据,以解决用户评分数据的稀疏性问题;将处理后的用户评分数据在基于正则化迭代最小二乘方法推荐系统中进行应用,取得良好效果.实验结果表明:使用Epinions数据集,相比传统的矩阵分解算法,该算法的平均绝对误差下降了10.77﹪.

Due to the data sparsity problems which lead to inaccuracy of recommendation in collaborative fil- tering recommender systems, a matrix factorization algorithm was proposed, which uses the trust propaga- tion. The proposed method used the direct trust relationship between users of social networks, based on the idea of trust propagation, predicted indirect trust relationship in social networks to solve the problem of sparse trust relations; the algorithm used social network trust data after filling for predicting and filling the user rating data to solve the problem of the sparsity of user rating data; It achieved good results in the recom- mender system based on the alternating - least - squares with weighted - λ - regularization method using user rating data that predicted and filled. The experimental results show that： the average absolute error of this algorithm decreases 10.77 % on the Epinions data sets, compared with the traditional algorithms.