基于多头注意力的社交网络用户身份链接方法  

Social network user identity linkage method based on multi-head attention

作　　者：臧文羽 颉夏青 邱莉榕 陆月明[2] Zang Wenyu;Xie Xiaqing;Qiu Lirong;Lu Yueming(Academy of Cyber,Beijing 100041,China;Key Laboratory of Trustworthy Distributed Computing and Service(BUPT),Ministry of Education,Beijing 100876,China)

机构地区：[1]网络空间研究院,北京100041 [2]北邮可信分布式计算与服务教育部重点实验室,北京100876

出　　处：《电子技术应用》2024年第12期61-64,共4页Application of Electronic Technique

基　　金：国家自然科学基金(62072488)。

摘　　要：随着社交网络的快速发展,人们在社交网络中拥有越来越多的虚拟身份,识别同一自然人不同网络虚拟身份的网络用户身份链接问题变得越来越重要。用户身份链接有助于挖掘网络用户的隐信息,构建全面的网络用户画像,进而促进跨网络的推荐、链接预测、信息传播等多个研究领域发展。现有的基于用户属性和基于网络结构的用户身份链接方法,没有考虑不同用户之间影响力差异因素,收敛速度较慢。基于深度游走的用户身份链接方法,融入多头注意力机制,对用户间影响力进行建模,实验结果表明,该方法可以很好地改进算法有效性,提高训练效率。With the rapid development of social networks,people have various virtual identities in social networks.User identity linkage problem that aims to identify various virtual identities of the same natural person is becoming increasingly important.User identity linkage method can unearth some hidden information and form a complete user profile to promote the development of multiple research fields,such as cross-network recommendation,link prediction,information dissemination,etc.Existing user-profile based model and network-structure based user identity linkage model do not consider the influence difference between dif-ferent users,and the convergence speed is slow.In order to model the influences between users,multi-head attention mechanism is added to network random-walk based user linkage method in this paper.The experimental results show that it can improve the effectiveness of social network user identity linkage method and improve training efficiency.

关 键 词：图表示学习 用户身份链接 多头注意力机制 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]