基于聚类信息和对称非负矩阵分解的链路预测模型研究

现有的大部分基于非负矩阵分解的链路预测方法仅考虑网络拓扑结构信息而忽略节点与链接聚类信息.针对此问题,提出一个融合聚类信息的对称非负矩阵分解的链路预测模型.首先,该模型利用对称非负矩阵分解去捕获网络节点相似度信息;其次,使用基于Jaccard的节点和链接聚类系数去保持网络局部结构信息;最后,启用拉格朗日乘法规则去学习模型参数.在六个真实无向无权和四个加权网络上的实验结果表明,该方法在两种不同类型网络预测精确度分别提升了1.6％和8.9％.

Link prediction model based on clustering information and symmetric non-negative matrix factorization

The goal of link prediction is to predict the possibility of forming links between unlinked nodes according to known network topology and node attributes. However, most existing link prediction methods based on non-negative matrix factorization only considered network topology information and ignore node and link clustering information. To solve this problem, this paper proposed a link prediction model based on symmetric non-negative matrix factorization with clustering information. Firstly, the model used symmetric non-negative matrix factorization to capture the similarity information of network nodes. Secondly, it used the node and link clustering coefficients based on Jaccard to keep the local structure information of the network. Finally, it enabled the lagrange multiplication rule to learn the model parameters. Experimental results on 6 real undirected unweighted and 4 weighted networks show that the prediction accuracy of this method on two different types of networks is improved by 1.6% and 8.9%, respectively.