摘要
质量控制是众包环境下一个极其关键的设计目标.传统策略的研究主要集中在如何通过对结果的质量评估、工作者的组织管理和众包任务的优化分配,达到质量控制的目的.但这些策略仍受到欺骗类型的工作者的影响,存在无法排除所有欺骗类型的工作者的问题,会导致结果的质量参差不齐.为了选取高质量的工作者,提高众包结果的质量,本文提出了一种基于动态选取工作者的质量控制模型(WST).将长时间的众包活动分为多个轮次,每个轮次分为两个阶段:离线阶段和在线阶段.离线阶段将已完成的任务使用聚类方法分为不同的同质簇,并计算每个簇与工作者的属性的关联度;在线阶段将发布的任务与已经存在的每个簇进行匹配,利用基于学习的模型为每个任务选取目前在线的最合适的工作者.实验研究了不同因素对准确率的影响,实验结果表明,与现有方法相比,WST模型能够取得更高的准确率,进一步验证了算法的有效性.
Quality control is an extremely critical design goal in a crow dsourcing environment.Traditional strategies mainly focus on the purpose of quality control by using the evaluation of the quality of the results,the organization of the w orkers and the optimal allocation of the crow dsourcing tasks.These quality control strategies are still affected by malicious w orkers,and it is impossible to exclude all malicious w orkers,resulting in uneven quality results.To select high-quality w orkers,this paper proposes a quality control model based on w orker selection(WST)to control the quality of crow dsourcing activities.This model divides long-term crow dsourcing activities into multiple rounds.Each round is divided into tw o phases,offline phase,and an online phase.In the offline phase,The completed tasks are divided into different homogeneous clusters using the clustering method,and the degree of association betw een each cluster and the w orker’s attributes is calculated.In the online phase,the published tasks are matched to each cluster that already exists.At the same time,a learning-based model w ill be used to select the most appropriate w orker currently online for each task.Experiments have examined the effects of different factors on the accuracy,The results show that the WST model can achieve higher accuracy than the existing methods,the effectiveness of the algorithm is further verified.
作者
高丽萍
金涛
GAO Li-ping;JIN Tao(School of Optical-Electrical Computer Engineering,University of Shanghai for Science&Technology,Shanghai 200093,China;Shanghai Key Laboratory of Data Science,Fudan Universit,Shanghai 200093,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2020年第10期2017-2023,共7页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61202376,61572325)资助
上海市自然科学基金项目(17ZR1429100)资助
上海市数据科学重点实验室重点开放课题项目(201609060003)资助。
关键词
众包
质量控制
工作者选取
聚类和学习
crow dsourcing
quality control
w orker selection
clustering and learning
