考虑结构与行为特征的水军群组检测算法  

Group spam detection algorithm considering structure and behavior characteristics

作　　者：张琪[1] 纪淑娟[1] 张文鹏 曹宁 李宁[1] Zhang Qi;Ji Shujuan;Zhang Wenpeng;Cao Ning;Li Ning(Shandong Provincial Key Laboratory of Wisdom Mine Information Technology,Shandong University of Science&Technology,Qingdao Shandong 266590,China)

机构地区：[1]山东科技大学山东省智慧矿山信息技术重点实验室,山东青岛266590

出　　处：《计算机应用研究》2022年第5期1374-1379,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(71772107,62072288)。

摘　　要：在线评论对用户的购买决策有重要的影响作用,部分卖方为提高自身信誉或贬低竞争对手的产品,通过雇佣大量水军有组织、有策略地撰写虚假评论来误导潜在消费者。为了检测这种有组织的水军群组,提出了一个综合考虑网络结构与评论者的行为特征水军群组检测算法。首先,根据评分和评论时间相关性得到评论者之间的紧密度,构建评论者关系图;其次,基于构建的评论者关系图,利用标签传播方法检测社区,得到候选群组集合;最后,复原候选群组对应的二部图,以对比可疑度为评估指标,在每个二部图上找到最终的造假者。基于真实数据集的实验结果证明了该算法的有效性。Online reviews play a significant role in users’purchasing decisions.In order to improve their reputation or degrade their competitors’products,some sellers employ a large numbers of review spammers to write fake reviews systematically and strategically to mislead potential consumers.In order to detect such organized spammer groups,this paper proposed a group spam detection algorithm that comprehensively considered the network structure and the behavior characteristics of reviewers.In implementation,this algorithm first obtained the closeness between reviewers based on the relevance of ratings and review time,and constructed a reviewer relationship graph.Secondly,based on the constructed reviewer relationship graph,it used label propagation method to detect the community and got a set of candidate groups.Finally,it restored the corresponding bipartite graphs of the candidate groups,and then found the final spammers on each bipartite graph by taking contrast suspiciousness as a metric.Experimental results based on real datasets demonstrate the effectiveness of the proposed algorithm.

关 键 词：水军群组 评论者关系图 标签传播 二部图 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]