Research On Anti-Disguised Attack Crowdsouring Platform Based On Genetic Rule Task Assignment Model

Crowdsourcing leverages the collective intelligence of the Internet masses to solve problems that are difficult to solve with traditional computers and has become an important part of the global Internet economy.Reducing costs and improving accuracy are two of the most important goals of crowdsourcing.To improve accuracy,crowdsourcing uses a redundancy-based strategy that assigns tasks to multiple workers and aggregates their answers to infer the true value.This process is known as truth value discovery.With the widespread use of crowdsourcing,various attack methods have emerged.While existing truth discovery methods can detect some simple attacks,today’s malicious workers have developed more intelligent and stealthy ways to disguise attacks.Not only can they combine their efforts to subvert the answer when the probability of success of the attack is high,but they can also disguise themselves to increase their weight when the probability of success of the attack is low.Such attacks are difficult to defend against existing truth discovery methods and greatly reduce the accuracy of the crowdsourcing platform.In addition,hiring workers in crowdsourcing platforms incurs costs,yet existing cost control methods only consider collecting as many worker answers as possible within the budget,and do not take into account that selecting malicious workers in scenarios with disguise attacks may compromise the quality of the integrated data,thus reducing the benefits.Therefore,in this paper,we systematically investigate the above problem using concepts and methods related to machine learning and deep learning.The main research is as follows:(1)A new defense framework against disguised attacks(Truth Discovery against Disguised Attack,TD-DA)is proposed for the problem of disguised attacks in crowdsourcing platforms.The framework mainly consists of two parts: task assignment and truth discovery.In the task assignment phase,the workers are assigned based on their performance on the golden task using a sigmoid function to quantify their aggressiveness and reliability.In addition,for the initial state in which workers are always assigned gold tasks,the probability of gold task assignment is estimated using Weighted Arithmetic Mean(WAM).In the truth discovery phase,the latest worker aggressiveness and reliability are taken into account,and truth aggregation and weight estimation are performed iteratively.Finally,the framework is experimentally verified to be effective in defending against masquerade attacks.(2)A Task Assignment based on Genetic Rules(TA-GR)model is proposed on the basis of anti-disguised attacks.The model first defines an optimization problem.And based on the labels of tasks in the previous chapter,worker reputation is defined and a benefit-cost model based on worker reputation is proposed.In addition,since the greedy algorithm can only obtain the local optimal solution when solving the constrained problem,this paper proposes a Greedy Repair Strategy with Cost Budget(GRS-CB)to obtain the global optimal solution of the objective function.Finally,it is shown experimentally that our proposed method is effective.(3)Based on the previous algorithm model,we construct a crowdsourcing management platform for anti-disguised attacks,the Shield Crowd platform.There are three modules:requester module,user module and administrator module.The requester module supports the registration and login of requester and the posting and viewing of tasks,while the user module supports the registration and login of individual users and the viewing of task completion.The administrator module provides administrator’s blocking management for platform users and view functions for tasks and system overview.The integration of algorithm model and system is realized.