基于知识图谱嵌入的涉诈网络链接补全和关键节点识别  

Link completion and key node identification of fraudulent network based on knowledge graph embedding

作　　者：李泽卿 黄诚 曾雨潼 冷涛[2,3] LI Ze-Qing;HUANG Cheng;ZENG Yu-Tong;LENG Tao(School of Cyber Science and Engineering,Sichuan University,Chengdu 610065,China;Sichuan Police College,Intelligent Policing Key Laboratory of Sichuan Province,Luzhou 646000,China;Chinese Academy of Sciences,Institute of Information Engineering,Beijing 100085,China)

机构地区：[1]四川大学网络空间安全学院,成都610065 [2]四川警察学院智能警务四川省重点实验室,泸州646000 [3]中国科学院信息工程研究所,北京100085

出　　处：《四川大学学报（自然科学版）》2024年第3期38-48,共11页Journal of Sichuan University(Natural Science Edition)

基　　金：智能警务四川省重点实验室开放课题(ZNJW2024KFZD003);国家重点研发项目(2021YFB3100500);四川省科技厅应用基础项目(2022NSFSC0752)。

摘　　要：涉诈网站作为网络诈骗的常见载体之一,在网络犯罪中扮演着平台内容提供者的重要角色.该形式的犯罪具有高度的团队性与合作性,涉诈网站在内的涉诈资产之间往往呈现出极强的关联.涉诈资产、涉诈团伙等共同构成了一个庞大的涉诈网络.虽然已有不少研究者针对涉诈网站识别开展了相关研究,但目前针对涉诈资产的关联性研究还相对较少.由于涉诈网络中节点的匿名性,导致直接获取涉诈资产相关的身份信息极为困难.警务人员往往难以快速准确的对涉诈网站进行溯源反制.本文基于本体论构建了细粒度的涉诈知识图谱,创新性地将知识图谱嵌入应用于涉诈网站溯源领域,将涉诈网络中的关系抽象为多维复空间上的旋转操作,并以知识图谱嵌入向量为依据,通过向量的空间相似性探求涉诈实体间关系网络的相似性,利用模型进行实体关系的补全;此外,本文创新性地对涉诈知识图谱中关系对涉诈团队身份的揭示程度进行量化,利用加权后的涉诈关系来优化特征向量中心性算法,以挖掘其中的关键线索节点.实验结果表明,在资产关系补全上本文使用的模型有着较高的准确率,在包含37866个实体的数据集上的HITS@10准确率达到了47%,效果领先于其他知识图谱嵌入模型.在后续案例中证明,本文设计的关键线索挖掘方法能够有效地对涉诈资产进行关联溯源,并取得了显著的成效.Fraudulent websites,as a common medium for online scams,play a significant role as providers of platform content in cybercrime.This form of criminal activity exhibits a high degree of teamwork and collaboration,with fraudulent websites often demonstrating strong interconnections among fraudulent assets.These assets,including fraudulent websites and associated criminal groups,collectively form a vast network of fraudulent.Although numerous researchers have conducted studies on identifying fraudulent websites,research on the correlation of fraudulent assets remains relatively scarce.Due to the anonymity of nodes within fraudulent networks,acquiring direct identity information related to fraudulent assets is exceedingly challenging for law enforcement personnel,making it difficult to trace and counter fraudulent websites accurately and promptly.This paper,based on an ontological framework,constructs a fine-grained knowledge graph of fraudulence and innovatively embeds knowledge graphs into the field of tracing fraudulent websites.It abstracts relationships within fraudulent networks as rotational operations in multidimensional complex spaces to model entities and relationships within the fraudulent knowledge graph.By utilizing knowledge graph embedding vectors,he model to complete entity relationships.Furthermore,this paper innovatively quantifies the degree of revelation of fraudulent team identities by relationships within the fraudulent knowledge graph.It optimizes centrality algorithms for feature vectors by utilizing weighted fraudulent relationships to unearth key clue nodes within it.Experimental results indicate that the proposed model exhibits a higher accuracy in com‐pleting asset relationships.On the dataset containing 37866 entities,the HITS@10 accuracy reached 47%,surpassing other knowledge graph embedding models in effectiveness.Subsequent case studies demonstrate that the key clue mining method designed in this paper can effectively trace the associations of fraudulent assets,thereby achieving signi

关 键 词：知识图谱嵌入 涉诈团伙 链接预测 关键节点识别 

分 类 号：TP309.1[自动化与计算机技术—计算机系统结构]