基于特征分析的推荐系统托攻击检测算法研究

Shilling Attack Detection Algorithm Based on Feature Extraction for Recommendation Systems

协同过滤是推荐系统中普遍使用的一种推荐技术,然而协同推荐系统很容易遭受恶意用户的攻击。攻击者通过向系统注入大量有规律的攻击用户信息,达到人为操纵推荐系统的目的。为了检测系统中存在的攻击用户,通过研究攻击用户信息的统计特征,提出了一种基于特征分析的攻击检测算法。试验结果表明,该算法具有更高的检测率,有效缓解了推荐系统遭受托攻击操纵的问题,确保了推荐系统的可靠性。

Collaborative filtering is one pupularly used recommendation technique in recommender systems. However,reco- mmender systems based on collaborative filtering are highly vulnerable to what have been termed ＂shilling＂ or ＂profile inj- ection＂ attacks.Attackers inject a number of organized biased ratings in order to manipulate recommendation systems.In or-der to detect the shilling attack users,this paper analyze the statistical features of attackers detailedly and then propose anattack detection algorithm based on the feature analysis.The experimental results show that the proposed attack detec- tion al-gorithm could achieve higher detection rate.Effectively alleviate the problem of the shilling attack to recommenda- tion syste-ms and ensure the reliability of the recommendation systems.