基于多种社交关系的概率矩阵分解推荐算法

随着社交网络的发展,社会化推荐算法得到普遍应用,现有的推荐算法往往只是将一种社交关系引入到推荐系统,但在现实社交网络中用户之间往往存在多种社交关系。基于多子网复合复杂网络模型,利用共享用户特征矩阵,提出了基于多关系社交网络的矩阵分解推荐算法。通过在Epinions数据集上的实验结果分析,准确率评价指标MAE、RMSE和NMAE分别提高了34%、27%和7%,由此可以证明,多关系社交网络的矩阵分解推荐算法能有效提高推荐准确率。

Matrix Decomposition Recommendation Algorithm Based on Multi-Relationship Social Network

With the development of social networks,social recommendation algorithms are widely used.Existing recommendation algorithms often only introduce one kind of social relationship into the recommendation system,but in reality there are multiple social relationships between users.Based on the multi-subnet composite complex network model and the shared user characteristic matrix,this paper proposes a matrix decomposition recommendation algorithm based on the multi-relational social network.Through the analysis of experimental results on the Epinions data set,the accuracy evaluation indexes MAE,RMSE and NMAE increased by 34%,27%and 7%respectively.This proves that the matrix factorization recommendation algorithm of multi-relational social networks can effectively improve the accuracy of recommendation.