结合元路径的属性重边异质网络嵌入方法  

Method of Attributed Multiple-edge Heterogeneous Network Embedding with Meta-path

作　　者：汤启友 张凤荔[1] 王瑞锦[1] 周志远 TANG Qi-you;ZHANG Feng-li;WANG Rui-jin;ZHOU Zhi-yuan(School of Information and Software Engineering,University of Electronic Science and Technology of China,Chengdu 610054,China)

机构地区：[1]电子科技大学信息与软件工程学院,成都610054

出　　处：《小型微型计算机系统》2022年第9期1958-1967,共10页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61802033,61472064,61602096)资助;四川省区域创新合作项目(2020YFQ0018)资助;四川省科技计划重点研发项目(2020YFG0475,2018GZ0087,2019YJ0543)资助;博士后基金项目(2018M643453)资助;广东省国家重点实验室项目(2017B030314131)资助;网络与数据安全四川省重点实验室开放课题项目(NDSMS201606)资助.

摘　　要：网络嵌入(network embedding)技术已被广泛地应用于图数据挖掘领域,并取得了良好的表现.然而当前针对异质网络的嵌入模型研究主要集中于拓扑图为简单图的网络,没有考虑对象之间的多重关系及其关系属性,这可能导致节点嵌入语义不完整、下游预测任务性能不高的问题.针对此问题,本文设计了一种结合元路径的属性重边异质网络嵌入(Attributed Multiple-Edge Heterogeneous Network Embedding with meta-path,AMEHNE)方法.该方法通过子网抽取、子网嵌入、嵌入融合等步骤将具有多重关系的网络节点嵌入到一个低维、实值的空间,同时还基于关系属性优化了元路径采样方式.实验结果表明,该方法优于传统的网络嵌入方法,能够对小样本节点在分类、聚类任务上有较大提升效果.Network embedding technology has been widely used in the field of graph data mining,and has achieved good performance.However,the current researches on heterogeneous network embedding model mainly focus on the network whose topology graph is a simple graph.And these researches do not consider the multiple relationships and relationship attributes between objects.So these methods may lead to the problem of incomplete semantics for node embedding and low performance of downstream prediction task.To solve this problem,this paper designs a method of attributed multiple-edge heterogeneous network embedding with meta-path(AMEHNE).This method embeds the network nodes with multiple relationships into a low dimensional,real value space.This method has the steps of subnet extraction,subnet embedding and embedding fusion,and optimizes the meta-path sampling method based on the relationship attributes.The experiments results show that the method is superior to the traditional network embedding method,and has a greater improvement effect on small sampling nodes in the classification and clustering tasks.

关 键 词：网络嵌入 异质网络 多重关系 关系属性 元路径 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]