作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (12): 192-199. doi: 10.19678/j.issn.1000-3428.0059636

• 移动互联与通信技术 • 上一篇    下一篇

基于传播时空特性的社交网络灰帽用户检测

何欢1, 朱焱1, 李春平2   

  1. 1. 西南交通大学 信息科学及技术学院, 成都 611756;
    2. 清华大学 软件学院, 北京 100091
  • 收稿日期:2020-10-01 修回日期:2020-12-02 发布日期:2020-12-09
  • 作者简介:何欢(1996-),女,硕士研究生,主研方向为社交网络、数据挖掘;朱焱(通信作者),教授、博士、博士生导师;李春平,副教授、博士。
  • 基金资助:
    四川省科技计划项目(2019YFSY0032)。

Grey Hat User Detection in Social Network Based on Spatiotemporal Characteristics of Diffusion

HE Huan1, ZHU Yan1, LI Chunping2   

  1. 1. School of Information Science and Technology, Southwest Jiaotong University, Chengdu 611756, China;
    2. School of Software, Tsinghua University, Beijing 100091, China
  • Received:2020-10-01 Revised:2020-12-02 Published:2020-12-09

摘要: 社交网络灰帽用户极易隐藏且类型多样,导致现有检测算法适用性较差。提出一种基于传播时空特性的社交网络检测算法。构建用户生成内容传播网络度量白帽和灰帽用户在传播空间上的不同特性,融合时空传播特性并调节权重比例以提高分类性能。实验结果表明,该算法能有效检测不同类型灰帽用户,与用户特征分析、社交网络链接分析、多视图融合等主流灰帽用户检测算法相比,其在CAVERLEE、CRESCI-15、CRESCI-17等多个数据集上的准确率及AUC值最高分别提升26.08%和30.54%。

关键词: 社交网络, 灰帽用户, 网络传播, 特征融合, 用户检测

Abstract: Grey hat users in social networks are diverse and good at disguising, which reduces the generability of detection algorithms.To address the problem, this paper proposes a social network detection mechanism based on the spatiotemporal characteristics of diffusion, Diffusion Spatio-Temporal Characteristics(DSTC).By using the characteristics of the diffusion time sequence of the content generated by white hat/grey hat users, a User Generated Content(UGC) diffusion network is constructed to measure the different characteristics of the white/grey hat users in the diffusion space.Based on the spatio-temporal characteristics of diffusion, a mechanism is designed for social network user detection, which integrates the spatio-temporal characteristics of diffusion and adjusts the weight ratio to improve classification performance.Experimental results show that compared with the mainstream grey hat user detection algorithms, which are based on user characteristic analysis, social network link analysis or multi-view fusion, the proposed algorithm exhibits a significant improvement in accuracy and AUC value, which is increased by up to 26.08% and 30.54% respectively on data sets like CAVERLEE, CRESCI-15, CRESCI-17, et al, indicating that DSTC can effectively detect different types of potential grey hat users.

Key words: social network, grey hat user, network diffusion, feature fusion, user detection

中图分类号: