A Study Of Large-scale Anonymous Collusion Attack And Defense Strategies For Crowdsourcing

The crowdsourcing model aims to generate group wisdom emergence through the independent work of a large number of Internet users to achieve the aggregation of group wisdom,which has been a great success in data tag collection,public opinion surveys,and sentiment analysis.However,there are some malicious Internet users in the platform,which derive a series of malicious behaviors such as random answers,answer plagiarism and malicious account registration in order to get more bounty with less/no labor.In crowdsourcing platforms,Internet users tend to handle tasks anonymously,so some collusion groups are mostly formed offline spontaneously,and it is difficult to form large-scale collusion directly.Therefore,malicious users often use third-party platforms in the form of an implicit social network to conspire,but in practice,due to the limitations of the social circle,it is still difficult to meet the interests of malicious users to maximize,so there is a proprietary conspiracy model platform,that is,a form of conspiracy in the form of online interaction with anonymous third-party users.Therefore,this paper investigates this conspiracy model from an offensive and defensive perspective as follows:1、A large-scale anonymous collusion framework for crowdsourcing.In order to break the limitation of anonymity of crowdsourcing system,this paper considers using cloud server to build a storage server for workers’ collusion data.The conspiracy workers can use the cloud server as a data channel to get the historical answers of conspiracy associates to submit tasks.To improve the security of workers in this conspiracy framework,a frequency control module is added to control the frequency of worker conspiracy.In this conspiracy framework,it is difficult to detect conspiracy workers in the system,whether it is tracking their IP addresses,filtering low-capacity workers,or analyzing changes in their capabilities,as will be confirmed by extensive experiments in this paper.Moreover,the colluding workers can get a multiplication of commissions for the same amount of work in this collusion.2、Neighborhood multiple weighted k-shell decomposition detection algorithm.Crowdsourcing tasks are issued redundantly,and a task is usually distributed to multiple online workers to complete,so there are a large number of cross-repeated tasks among workers.Based on this this paper defines the worker answer consistency ratio to portray the collusion weights among workers.Based on the idea of k-shell algorithm this paper maps the worker population into a graph network and identifies colluding workers by analyzing the density distribution of workers in the graph with collusion weights.This detection algorithm can detect the small and large groups involved in collusion among a large number of crowdsourced workers.This paper demonstrates that the detection algorithm is superior to some existing detection algorithms through experiments in multiple dimensions.