基于CBOW模型的链路预测方法  

Link Prediction Method Based on CBOW Model

作　　者：赵宇红[1] 张政 ZHAO Yu-hong;ZHANG Zheng(School of Information Engineering,Inner Mongolia University of Science and Technology,Baotou 014010,China)

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014010

出　　处：《小型微型计算机系统》2020年第11期2392-2398,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61562056)资助;内蒙古自然科学基金项目(2016MS0608)资助.

摘　　要：为了有效、准确地挖掘节点自身的属性与网络结构的关系信息并将其联合应用于链路预测,受概率语言检索研究的启发,提出基于CBOW模型的链路预测方法.通过使用包含节点邻居信息和网络连通信息的节点序列库训练CBOW模型产生节点向量,结合节点向量自身属性和节点对之间的趋向程度提出一种新的相似性评价指标—向量自量趋向性(SMTV),使用此相似性指标进行网络链路预测.在PPI-Yeast、Facebook和Power Grid三个真实数据集上进行实验,分别对比CN,AA,LP和Node2vec-Hadamard四种方法的AUC值,CBOW-SMTV相比其中AUC最低的方法,分别有5.3109%、14.4955%、41.9747%的提高;相比AUC最高的方法也有0.2497%、0.6921%、9.5714%的提高.因此基于CBOW-SMTV的链路预测方法能有效结合节点属性和网络结构信息,提高链路预测有效性.It has become an important mission to learn about the effective and precise data mining of how intrinsic nodal features are related with the network structure,and to understand how this data mining can be applied in combination with link prediction.In light of this background,this paper draw inspiration from the research on probabilistic language retrieval and propose a CBOW-model-based link prediction method.The CBOW model is trained to generate node vectors using the node sequence library,which contains node neighbor information and network connectivity information.A new similarity evaluation index,i.e.,self-measurement tendency of vector(SMTV),is proposed based on the attributes of node vectors and the trend degree of node pairs,and then used for the network link prediction.The AUC values of CN,AA,LP and Node2vec-Hadamard are compared respectively on the three real datasets of PPI-Yeast,Facebook,and Power Grid.Compared with the lowest AUC method,CBOW-SMTV showed 5.3109%,14.4955%,and 41.9747%improvement;compared with the method with the highest AUC,0.2497%,0.6921%,and 9.5714%improvement.Therefore,the CBOW-SMTV-based link prediction method can effectively combine node attributes and network structure information to improve the effectiveness of link prediction.

关 键 词：复杂网络 信息挖掘 CBOW模型 链路预测 相似性指标 AUC 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]