基于图核的异质信息网络链路预测方法  

Graph kernel based link prediction in heterogeneous information network

作　　者：赵妍[1,2,3] 赵书良 马秋微[1,2,3] Zhao Yan;Zhao Shuliang;Ma Qiuwei(College of Computer&Cyber Security,Hebei Normal University,Shijiazhuang 050024,China;Hebei Provincial Engineering Research Center for Supply Chain Big Data Analytics&Data Security,Shijiazhuang 050024,China;Hebei Provincial Key Laboratory of Network&Information Security,Shijiazhuang 050024,China)

机构地区：[1]河北师范大学计算机与网络空间安全学院,石家庄050024 [2]供应链大数据分析与数据安全河北省工程研究中心,石家庄050024 [3]河北省网络与信息安全重点实验室,石家庄050024

出　　处：《计算机应用研究》2021年第10期3125-3130,共6页Application Research of Computers

基　　金：国家社会科学基金重大项目(13&ZD091,18ZDA200);河北省重点研发计划资助项目(20370301D)。

摘　　要：链路预测是图挖掘主要研究的问题,其研究重点是提取图的特征信息,现有研究方法大多只关注网络拓扑结构而忽略了节点属性信息。针对该问题,提出了基于图核的链路预测方法NGLP。该方法能挖掘有效、可用的元路径;基于元路径对预测对象生成带节点属性的子图,使用子图表示被预测的链路;然后利用图核方法计算子图之间的相似性;最后训练SVM得出链路预测结果。实验结果表明,提出方法与其他方法相比具有更高的精度和更强的稳定性。Link prediction is the main research problem of graph mining,which focuses on extracting feature information from the graph.Most of the existing research methods only focus on the network topology and ignore the node attribute information.To address this problem,this paper proposed the graph kernel-based link prediction method(NGLP).Firstly,it mined the valid and usable meta-paths.Secondly,it generated subgraphs with node attributes for the predicted objects based on meta-paths,and used the subgraphs to represent the links being predicted.Then,it used the graph kernel method to calculate the similarity between subgraphs.Finally,it trained the SVM to produce link predictions.Experimental results show that the proposed method has higher accuracy and more excellent stability compared to other methods.

关 键 词：异质信息网络 链路预测 图核 元路径 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]