基于关系的跨网络用户身份链接  

Relationship-based user identity linkage across networks

作　　者：刘红[1] 朱焱[1] 李春平[2] LIU Hong;ZHU Yan;LI Chun-ping(School of Computing and Artificial Intelligence,Southwest Jiaotong University,Chengdu 611756,China;School of Software,Tsinghua University,Beijing 100091,China)

机构地区：[1]西南交通大学计算机与人工智能学院,四川成都611756 [2]清华大学软件学院,北京100091

出　　处：《计算机工程与设计》2023年第6期1649-1655,共7页Computer Engineering and Design

基　　金：四川省科技计划基金项目(2019YFSY0032)。

摘　　要：为打破现存研究普遍以网络拓扑一致性假设为前提的限制,弱化锚节点数量和质量对链接任务的影响,提出一种基于跨网络语义表征的用户链接算法CSRMA(cross-network semantic representation link algorithm based on mask attention mechanism)。该算法框架包含3个模块:多视角采样与注意力机制相结合的跨网络表征模块、不同网络共性特征学习的语义空间映射模块、基于k-d树改进Gale-Shapley算法的用户身份精准链接模块。通过4个公开数据集上的实验验证了所提算法的有效性。与多个身份链接算法对比,CSRMA具有更高的精确率。To break the limitation of existing researches based on the assumption of network topology consistency and weaken the influence of the number and quality of anchor nodes on the task of linkage,a user linking algorithm based on cross-network semantic representation(CSRMA)was proposed.The algorithm framework consisted of three modules including the cross-network representation module combining multi-perspective sampling and attention mechanism,the semantic space mapping module of different network common features learning,and the user identity precise link module based on k-d tree improved Gale-Shapley algorithm.The proposed algorithm was validated by experiments on four public datasets.At the same time,CSRMA has higher precision than multiple identity linking algorithms.

关 键 词：用户身份链接 跨网络 用户关系 采样策略 共性特征 掩码注意力机制 精准链接 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]