多尺度特征融合的托攻击检测方法  

Shilling attack detection method based on multi-scale feature fusion

作　　者：于金霞[1] 李佳昕 李星宇 汤永利[2] YU Jinxia;LI Jiaxin;LI Xingyu;TANG Yongli(School of Computer Science and Technology,Henan Polytechnic University,Jiaozuo 454003,P.R.China;School of Software,Henan Polytechnic University,Jiaozuo 454003,P.R.China)

机构地区：[1]河南理工大学计算机科学与技术学院,河南焦作454003 [2]河南理工大学软件学院,河南焦作454003

出　　处：《重庆邮电大学学报（自然科学版）》2023年第5期863-872,共10页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：河南省高校科技创新团队支持计划(20IRTSTHN013);河南省高校基本科研业务费专项资金项目(NSFRF210312);河南理工大学博士基金项目(B2021-41)。

摘　　要：针对利用先验知识和人工特征提取技术检测托攻击方法中存在的灵活性差、稳定性低问题,提出一种基于多尺度特征融合的托攻击检测方法。使用图像可视化技术将评分行为以用户为单位转化为标准的灰度粒度图像,提高对托攻击的检测灵活性;构建多尺度特征融合神经网络检测模型,该模型利用特征提取模块分别从标准的灰度粒度图像中提取出可以表示局部评分信息的大尺度特征图层和表示整体评分信息的小尺度特征图层,通过特征融合模块整合不同尺度特征图层的信息,进而提升托攻击检测的稳定性。实验结果表明,在不同填充率和不同攻击规模下,该方案对托攻击的检测精确率和准确率均达到90%以上,对托攻击的检测具有良好的灵活性和稳定性。Aiming at the problems of poor flexibility and low stability in the method of detecting shilling attack using prior knowledge and artificial feature extraction technology,a shilling attack detection method based on multi-scale feature fusion is proposed.Image visualization technology is used to transform the rating behavior into the standard gray-scale image on a user basis to improve the flexibility of detecting the shilling attack.We construct multi-scale feature fusion neural network detection model.The feature extraction module of this model can extract large-scale feature layers which can represent local rating information and small-scale feature layers which can represent overall rating information from the standard gray-scale image.The feature fusion module of this model can integrate the information of feature layers of different scales to improve the detection stability of shilling attack.The experimental results show that the scheme can detect more than 90%of the shilling attack accuracy rate and precision rate under different filling rates and attack scales.The proposed shilling attack detection method has good flexibility and stability.

关 键 词：托攻击检测 神经网络 多尺度特征融合 

分 类 号：TN919.8[电子电信—通信与信息系统] TP391.3[电子电信—信息与通信工程]