摘要
移动互联网技术的飞速发展,给社交网络平台带来了新的颠覆性的转变,也不断地改变着人们的生产、生活和交流方式.在线社交网络由于其特有的注册开放性、发布信息自由性、用户兴趣趋同性等特点,已经超越传统媒体,成为人们传播消息、获取新闻和接收实时信息的主要途径.同时,社交网络中用户之间的各种关系类型多样、相互交织、相互影响,促使用户生活在复杂的在线群体网络环境中,使得用户的在线行为时刻都受到所属的多种群体环境的影响作用.现有的针对在线群体环境影响的研究大多依据静态的、单一的网络结构对社交网络进行建模,而网络中通常存在多种类型的、动态的社会关系,较少研究能同时考虑多种类型的用户关系,建模社交网络中复杂环境下用户受到的影响作用.本文对用户所处的多类在线群体环境进行分析,挖掘用户所能感知的不同类型的群体环境,建模多维群体环境下用户所受的影响作用.首先,从用户间的社交关系类型出发,对在线社交网络中复杂的网络拓扑关系进行分类挖掘,分析用户可能感知的不同维度的在线群体环境,并提出静态群体环境和动态群体环境的定义和挖掘方法.其次,在不同的在线社交群体环境下,从宏观角度量化环境中用户所感知的群体结构特征,并从微观角度建模并模拟用户间的影响机制,提出了基于图注意力网络的融合多维在线群体环境的影响力模型.最后,以在线社交网络中用户的转发行为为例,研究多维群体环境影响下的用户行为模式,并在真实数据集上,基于群体影响力模型预测个体转发行为状态,验证模型的合理性和有效性.实验结果表明,本文提出的群体影响力模型能够更有效地描述在线社交网络中用户所属群体对用户的影响作用,并且在用户转发行为状态预测方面,比现有的群体影响力模型在综合评价指标F1值方面最高可以提升33%,在AUC值方面可提升16%.
The rapid development of Mobile Internet technology has brought new and subversive changes to online social network platforms,and also has changed the styles of people’s production,lifestyle,and the way of communication.The characteristics of online social networks(OSNs),such as the openness of registration,the freedom of information diffusion,and the homophily of users’interests,have made OSNs overtaken traditional media(e.g.newspapers,TV,magazines)as one of the main channels for users to reading news,show their daily life and receive their friends’dynamic information.At the same time,various types of relationships among users in social networks are interwoven and influence each other,which makes the online social network environments more complex and informative.Furthermore,users’online social behaviors are also affected by the diverse online social group environments that they belong to,which brings us a lot of challenges to social influence analysis.Recently,most of the previous studies that focus on the social influence of online group environment only use single type of relationship and assume that the relationships among users are static,and few studies can consider diverse types of relationships and the dynamics at the same time when quantifying the influence impacted on users in the complex environment of the social network.In this paper,we mine the diverse types of group environments that users perceive in OSNs and model the influence of users’multi-relational group environments.At last,we evaluate the influence of the online group environment that impacted on individuals and propose a Multi-Relational Group Influence model(MRINF)to predict the status of users’online retweet behaviors.Specifically,First,we analyze the diverse types of online social relationships and mine structural characteristics of complex network in online social networks platforms.Then,we give a deep analysis of users’perceived social group environment and propose two formalized definitions and give the group detection methods separately to find the potential group environments in social networks,which include the static social group environment and the dynamic social group environment.Secondly,we quantify the macroscopic structural features of two types of users’perceived groups.Moreover,using convolutional operation in graph attention network to simulate the information diffusion in OSNs,a multi-relational group influence model is proposed that combines the microscopic influence process among users and the macroscopic perceptions features.Finally,taking users’retweet behavior status as the main explicit expression of group environment influence in social networks,the proposed model is applied to the application of predicting individual retweet behaviors on two datasets and compared with the state-of-the art algorithms to verify the rationality and effectiveness of existing models.The extensive experimental results show that the MRINF model that is proposed in this paper could effectively describe the influence of users’perceived groups that impact individuals in online social networks from the static and dynamic perspective.In terms of the task of users’retweet behavior prediction,the MRINF outperforms other state-of-the-art algorithms on the evaluation metrics.Specifically,our model has an improvement of over 33%in F1-value and 16%higher in Area Under Curve(AUC)value compared with the existing state-of-the-art social influence model.
作者
孟青
刘波
张恒远
孙相国
曹玖新
李嘉伟
MENG Qing;LIU Bo;ZHANG Heng-Yuan;SUN Xiang-Guo;CAO Jiu-Xin;LEE Roy Ka-Wei(School of Computer Science and Engineering,Southeast University,Nanjing 211189;School of Cyber Science and Engineering,Southeast University,Nanjing 211189;Department of Computer Science,University of Saskatchewan,Saskatoon,Canada,S7N 5C9)
出处
《计算机学报》
EI
CAS
CSCD
北大核心
2021年第6期1064-1079,共16页
Chinese Journal of Computers
基金
国家重点研发计划项目(2017YFB1003000,2019YFC1521403)
国家自然科学基金项目(61972087,61772133,61632008)
国家社会科学基金项目(19@ZH014)
江苏省自然科学基金项目(SBK2019022870)
江苏省网络与信息安全重点实验室(BM2003201)
江苏省计算机网络技术重点实验室(BE2018706)
教育部计算机网络与信息集成重点实验室(东南大学)(93K-9)资助.
关键词
在线社交网络
群体
群体影响力
环境感知
图注意力网络
online social network
groups
group influence
environment-aware
graph attention networks
