基于模块度相似性的二分网络链路预测算法

Link Prediction in Bipartite Networks Based on Modularity Similarity

基于社团结构,提出模块度相似性的二分网络链路预测算法,克服了二分网络在链路预测中丢失社团结构信息的局限性。首先,通过定义二分模块度,利用奇异值分解,将网络中的节点嵌入到欧式空间中的向量。其次,提出二分网络模块度相似性的框架,利用向量余弦相似度定义二分网络节点对之间的模块度相似性指标(MS指标)。最后,基于小提琴图和评价指标AUC,在3个真实网络上进行模拟仿真,与9种链路预测相似性指标进行对比,证明MS指标用于二分网络链路预测具有较高的精度。

In this paper, a bipartite network link prediction algorithm for modularity similarity is proposed based on community structure. The limitations of bipartite networks in link prediction that lose community structure information are overcome. First, by defining the bipartite modularity and using singular value decomposition, the nodes in the network are embedded into vectors in the Euclidean space. Secondly,a framework for bipartite networks modularity similarity is proposed. The modularity similarity metrics(MS metrics) between pairs of nodes in the bipartite network are defined using vector cosine similarity. Finally, simulations are performed on three real networks based on violin graph and evaluation metrics AUC.The MS metrics are compared with nine link prediction similarity metrics. The MS metrics are used for bipartite networks link prediction with high accuracy.