基于多维动态特征验证的社交机器人账号检测  

Social robot account detection based on multi-dimensional dynamic feature verification

作　　者：胡舫舷 罗文华[1] HU Fang-xian;LUO Wen-hua(College of Public Security Information Technology and Information,Criminal Investigation Police University of China,Shenyang 110035,China)

机构地区：[1]中国刑事警察学院公安信息技术与情报学院,辽宁沈阳110035

出　　处：《佛山科学技术学院学报（自然科学版）》2023年第1期23-34,共12页Journal of Foshan University(Natural Science Edition)

基　　金：辽宁省社会科学规划基金重点项目(L21AFX006);中国刑警学院研究生创新能力提升项目(2021YCYB43)。

摘　　要：社交机器人账号的存在增加了无关、无效甚至虚假信息的传播机率,影响人类的线上及线下生活。人工识别此类账号需要工作者经验丰富,同时成本高、精度低。从账号动态变化角度出发引入社交关系关注数变化量、粉丝变化率和基于序列匹配的发文相似度特征,建立多维动态特征社交机器人账号检测模型,提出一种改进的基于AUC决策树分类评价指标随机森林优化算法以期解决以上问题。实验结果表明,改进后的算法对于提升检测准确率具有明显帮助,在社交机器人检测更注重的查全率方面也具有一定的优势。The existence of social robot accounts increases the spread probability of irrelevant,invalid and even false information,affecting people’s online and offline life.Manual identification of such accounts requires rich experience,high cost and low precision.From the perspective of dynamic change of accounts,social relations focus number variation,fans rate and posts based on sequence matching similarity characteristics are introduced to establish a multi-dimensional dynamic feature social robot account detection model.An improved decision tree classification evaluation index based on the AUC random forest optimization algorithm is proposed to solve the above problem.Experimental results show that the improved algorithm has significant help to improve the detection accuracy,and has certain advantages in the aspect of recall rate,which is more important in social robot detection.

关 键 词：社交机器人 特征验证 随机森林 AUC值 决策树 查全率 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]