Research On Quality Control Based On Reputation Model In Crowdsourcing

In this era of everything being interconnected,crowdsourcing,as a new type of distributed problem-solving approach,has developed rapidly.In the many studies on crowdsourcing,a key issue is quality control.Current quality control methods can generate a large number of malicious workers and rarely consider the impact of workers providing trustworthy services multiple times on the quality of results.Existing worker selection mechanisms mostly meet the needs of requesters by considering cost,location,and reputation indicators,and by ignoring requester bias mechanisms based on worker preferences,the motivation of workers to complete tasks is low,ultimately affecting the satisfaction of both parties.In addition,workers may lose interest in the tasks they are matched with,which can lead to unstable task-worker matching.In response to these issues,this paper conducts research with the following main content:(1)A reputation-based crowdsourcing quality control method is proposed.First,a reputation model is established based on trustworthy and punishment factors.Then,a worker selection mechanism is proposed based on worker reputation values and familiarity with tasks.At the same time,the worker matching degree,which combines reputation values and familiarity,is used as a weight and introduced into the EM(Expectation Maximum)algorithm,which is called the Qrep-EM algorithm.This solves the problem of EM algorithm sensitivity to initial values and convergence difficulties,avoids the algorithm falling into local optima,and improves the accuracy of evaluation results.Finally,the Qrep-EM algorithm and the mechanism proposed in this paper are verified using the publicly available crowdsourcing datasets Adult2 and Duck.The experimental results show that the Qrep-EM algorithm has significantly improved evaluation accuracy and running time compared to the comparison algorithm,and the effectiveness of the worker selection mechanism proposed in this paper is also verified from the perspectives of task completion rate and average data quality.(2)A dynamic reputation update strategy based on Deep Q-learning is proposed.Based on the current state,including the previous reputation strategy,average number of completed tasks,and average reputation score of malicious workers,an action is selected to maximize the reward,and then the environment is transitioned to a new one,repeated multiple times until the optimal strategy,i.e.,the reputation update strategy,is determined.The Qo I and Qo T indicators are used to evaluate workers and tasks,and a preference list of workers and tasks is formed based on this.Based on the traditional Gale-Shapley stable matching algorithm,a worker matching mechanism that considers both requester and worker preferences is proposed,which not only improves the quality of workers selected by requesters but also increases the satisfaction of workers performing tasks.Finally,experiments are conducted on publicly available datasets,and the results show that the proposed method performs better.