一种基于卷积自动编码器的推荐系统攻击检测方法

Recommendation System Attack Detection Method Based on Convolutional Autoencoder

协同过滤推荐已经成为解决互联网上信息过载的有效方法之一,但是,协同过滤推荐系统本身所具有的高度开放性,容易受到恶意用户的攻击,导致产生欺诈性的推荐结果,因此,有效的攻击检测对于提高推荐系统的可用性具有重要的意义.特征工程的质量很大程度上决定了攻击检测性能,而目前大多数攻击检测方法都是基于人工方式来提取用户特征,面对不同的攻击模型,构建通用的、合适的特征指标往往是非常困难的,因此,本文提出了一种基于卷积自动编码器的推荐系统攻击检测方法,将自动特征提取和人工设计特征相结合来构造攻击检测特征,将自动编码器与卷积神经网络相结合,以卷积神经网络的卷积操作完成自动编码器的编码和解码功能,实现特征自动提取,采用深度学习方法进行攻击检测.实验验证了本文提出方法的有效性.

Collaborative filtering recommendation has become one of the effective methods to solve the Internet information overload,but the collaborative filtering recommendation system itself is highly open and vulnerable to malicious users,lead to fraudulent recommended as a result,therefore,effective attack detection to improve the usability of recommendation system has important significance.Characteristics of engineering largely determines the quality of the attack detection performance,and most current attack detection methods are based on the artificial way to extract the user characteristics,in the face of different attack model,the construction of a general,suitable characteristic index is often very difficult,therefore,this paper proposes a recommendation system based on convolutional autoencoder automatically attack detection method,combining the automatic feature extraction and artificial design features to construct the attack detection characteristics,combining autoencoder and convolutional neural network,The function of autoencoder encoding and decoding is accomplished by convolution operation of convolutional neural network,to realize automatic feature extraction,Deep learning method is used to detect attacks.Experiments verify the effectiveness of the proposed method.