基于信誉模型的众包质量控制算法  

Crowdsourcing quality control algorithm based on reputation model

作　　者：褚佳静 潘庆先 潘亚楠 刘庆菊 CHU Jiajing;PAN Qingxian;PAN Ya'nan;LIU Qingju(School of Computer and Control Engineering,Yantai University,Yantai 264005,Shandong,China)

机构地区：[1]烟台大学计算机与控制工程学院,山东烟台264005

出　　处：《山东大学学报（工学版）》2023年第2期93-101,共9页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金项目(60903098,61502140,61572418,61472095,62072392)。

摘　　要：针对目前众包平台会产生大量恶意工人以及较少考虑激励工人多次提供可信服务的问题,提出一种基于信誉模型的众包质量控制算法——信誉期望最大化(reputation expectation maximum,Rep-EM)算法。根据可信因子和惩罚因子建立信誉模型;基于工人信誉值和对任务的熟悉度提出一种工人选择机制;将工人匹配度作为权重赋予相应的工人并使用多数投票方法进行初始值选取,解决期望最大化(expectation maximum,EM)算法对初始值敏感和收敛困难的问题,避免算法陷入局部最优,提高评估结果的准确率;利用公开的众包数据集Adult2和Duck对Rep-EM算法和本研究提出的机制进行验证。试验结果表明,Rep-EM算法在评估准确率和运行时间方面有很大的提升,也从任务完成率和平均数据质量验证了本研究提出的工人选择机制的有效性。Towarding the problems that the current crowdsourcing platform produced a large number of malicious workers and less consider encouraging workers to provide trusted services many times,a crowdsourcing quality control algorithm(reputation expecta-tion maximum,Rep-EM)was proposed based on reputation model.The reputation model was constructed based on the credibility factor and penalty factor.A worker selection mechanism was proposed based on workers'reputation and familiarity with tasks.The matching degree of worker was given to the corresponding workers as a weight,and the majority voting method was used to select the initial value,which solved the problem that the expectation maximization(EM)algorithm was sensitive to the initial value and difficult to converge.It could also avoid the algorithm falling into local optimization,and improve the accuracy of the evaluation re-sults.The Rep-EM algorithm and the mechanism were verified with the public crowdsourcing datasets Adult2 and Duck.The experi-mental results showed that the Rep-EM algorithm performed better in the evaluation accuracy and running time compared with other algorithm.The effectiveness of the worker selection mechanism was verified from the task completion rate and average data quality.

关 键 词：众包 质量控制 信誉模型 工人选择机制 EM算法 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]