结合注意力机制的属性异质网络嵌入方法  

Attribute Heterogeneous Network Embedding Method Combining Attention Mechanisms

作　　者：李嘉坤 王瑞锦[1] 张凤荔[1] 李冬芬 孙永佼 应时[4] LI Jiakun;WANG Ruijin;ZHANG Fengli;LI Dongfen;SUN Yongjiao;YING Shi(School of Information and Software Engineering,University of Electronic Science and Technology of China,Chengdu 610054,China;College of Computer Science and Cyber Security(Oxford Brookes College),Chengdu University of Technology,Chengdu 610059,China;College of Computer Science and Engineering,Northeastern University,Shenyang 110819,China;School of Computing,Wuhan University,Wuhan 430072,China)

机构地区：[1]电子科技大学信息与软件工程学院,成都610054 [2]成都理工大学计算机与网络安全学院,成都610059 [3]东北大学计算机科学与工程学院,沈阳110819 [4]武汉大学计算机学院,武汉430072

出　　处：《小型微型计算机系统》2024年第6期1466-1473,共8页Journal of Chinese Computer Systems

基　　金：国家重点研发计划项目(2022YFB3304300)资助。

摘　　要：图(网络)是一种常用于抽象现实世界实体之间关系的数据结构,网络嵌入广泛应用于图数据的表征.目前大部分异质网络嵌入方法未考虑网络节点之间的多种边类型和边属性,无法完整刻画网络的结构和语义信息,导致原始网络特征信息丢失和下游任务效果差的问题.为解决该问题,基于注意力机制设计了一种多边属性异质网络嵌入方法,其将注意力机制应用于学习不同边类型下嵌入向量的重要系数,通过有偏序列采样、邻居向量聚合和模型参数更新3个阶段的嵌入学习,将网络节点表示成固定长度的稠密向量.实验表明,提出的嵌入方法能够更好地嵌入网络的特征信息,使之在下游的机器学习任务上有一定的效果提升.Graph(network)is a kind of data structure commonly used to abstract the relationship between real-world entities,and network embedding is widely used to characterize graph data.Most current heterogeneous network embedding methods do not consider multiple edge types and edge attributes among network nodes,which cannot completely portray the structural and semantic information of the network,leading to the problems of loss of original network feature information and poor results in downstream tasks.To solve this problem,a multilateral attribute heterogeneous network embedding method is designed based on the attention mechanism,which applies the attention mechanism to learn the importance coefficients of embedding vectors under different edge types,and represents the network nodes as dense vectors of fixed length through three stages of embedding learning:biased sequence sampling,neighbor vector aggregation and model parameter updating.Experiments show that the proposed embedding method can better embed the feature information of the network,which makes it effective for downstream machine learning tasks to be improved.

关 键 词：表示学习 注意力机制 异质网络 图嵌入 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]