Research On Detecting Attack-Block Algorithms In Recommender System

Recommender system has been widely used in many areas such as e-commerce and social networks, as an efficient tool for information filtering. However cur-rently, in order to gain benefit, some malicious users called spammers try to influence the results of recommending by forging lots of fake users’profiles, which damages the benefit of businessmen and customers. And this behavior is called malicious attack. In general, there are two kinds of malicious attacks. One is "promoting attack", in which spammers generate lots of high-rated records to some specific items to promote them. The other is "nuke attack", in which spam-mers generate lots of low-rated records to some specific items to demote them. With the development of recommender system, detecting "malicious attack" has become a new research hotspot, and what we focus is just the "nuke attack" in "malicious attack".Many existing methods mainly have attempted to solve the problem of nuke attack from two perspective, they either focus on detecting individual spammers, neglecting the cooperative relationship between them, or try to firstly find the candidates of a group of spammers and attacked items separately, and detect the attack after. This could be not direct and inefficient, and can not handle the big data. Due to the lack of work aiming at simultaneously finding spammer groups and item groups, from a new perspective, this paper proposes an algorithm named MAB (Mining Attack-Block), which considers the spammer groups and attacked item groups simultaneously and uses the idea of frequent pattern mining to detect the attack behaviors. Besides, to face the challenges of big data, we exploit some pruning strategies to improve the efficiency of the proposed algorithm, and validate its effectiveness and efficiency on datasets. Our contributions can be listed as follows:1. To detect spammer groups and attacked item groups in "nuke attack" ef-ficiently, we propose to regard the relationship between "spammers" and the relationship between "spammers and attacked items" as one important feature of detecting attack behaviors. We also define the concept of Attack-Block based on it.2. Proposing two measures namely Block-Area-Ratio and Block-Rate-Ratio to detect Attack-Block. Then we propose an Attack-Block detection algorithm MAB based on the measures, this algorithm adopts ideas of frequent pattern mining to detect the Attack-Block. We exploit the upper bounds of these two measures to further prune the search space to improve efficiency of detection.3. Finally, we validate the MAB algorithm on datasets, experimental results show that MAB can detect Attack-Blocks correctly, and the pruning strategy that MAB used has a good performance, which leads to broad application prospects.