基于多视图证据融合的社交水军检测  

Social spammer detection based on multi-view evidence fusion

作　　者：张东林 徐建[1] Zhang Donglin;Xu Jian(School of Computer Science&Engineering,Nanjing University of Science&Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094

出　　处：《计算机应用研究》2024年第10期2939-2946,共8页Application Research of Computers

基　　金：国防基础科研计划国防科技重点实验室稳定支持项目。

摘　　要：为克服单视图水军检测方法在处理复杂多样的社交网络数据时所存在的局限性,并解决现有多视图融合方法未能充分考虑视图间质量差异导致的信息丢失和噪声干扰等问题,提出一种基于多视图证据融合(multi-view evidence fusion,MVEF)的社交水军检测方法。该方法综合分析社交关系、行为特征和推文内容三个视图并提取关键证据,通过Dirichlet分布参数化来评估每个视图在分类决策中的类别可信度和整体不确定性。通过高效的证据融合机制,巧妙地利用不确定性整合各视图中的关键证据,构建一个全面而可靠的分类决策框架。实验结果显示,MVEF在两个真实世界的Twitter数据集上的表现均优于现有方法,有效提升了水军识别的准确率和鲁棒性。To address the limitations of single-view spammer detection methods in processing complex and diverse social network data,and the issues of information loss and noise interference due to existing multi-view fusion methods not fully consi-dering the quality differences between views,this paper proposed a social spammer detection method based on MVEF.The method integrated and analyzed three views:social relationships,behavioral characteristics,and tweet content,to extract pi-votal evidence.It employed Dirichlet distribution parameterization to assess the category credibility and overall uncertainty of each view in classification decisions.Through an efficient evidence fusion mechanism,the method skillfully utilized uncertainty to integrate key evidence from various views,constructing a comprehensive and reliable classification decision framework.Experimental results demonstrate that MVEF outperforms existing methods on two real-world Twitter datasets,effectively enhancing the accuracy and robustness of spammer detection.

关 键 词：社交水军检测 多视图 证据融合 不确定性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]