混杂协作式环境中基于多Agent博弈的众包质量优化模型研究

For hybrid crowdsourcing systems including non-collaborative and collaborative workers, this proposed research applies multiagent game-theoretic techniques to model the motivation behind the non-collaborative and collaborative behaviors. The proposed research breaks through the limitations of traditional single-type crowdsourcing systems including only non-collaborative workers. The proposed research studies the effect of the hybrid crowdsourcing setting on task completion quality, and explores adequate crowdsourcing models, payoff mechanisms and task allocation mechanisms. Specially, the proposed research a) presents the team formation-based game-theoretic crowdsourcing model, which can encourage multiagents to complete tasks collaboratively and can reveal the advantages of collaboration on task completion quality, b) presents the spot-checking mechanism, which can prevent multiagents’ malicious collaboration behavior and elicit multiagents to provide meaningful and valuable task completion results, and c) presents the social norm-based dynamic evolving crowdsourcing model with hybrid non-collaborative and collaborative agents, which can reveal the necessary and sufficient conditions of convergence from non-collaborative systems to collaborative systems. .The results of the proposed research can explore the essence of the hybrid crowdsourcing systems from the multiagent game-theoretic perception. It can also help to solve the task completion quality issue of the massive collective intelligence-based crowdsourcing systems.

本项目针对众包环境中工作者非协作性与协作性混杂行为特征，采用多Agent博弈方法对工作者非协作与协作行为背后的理性动机进行建模，从而突破传统单一的针对非协作式工作者的众包质量优化模型的局限性。研究混杂协作式众包环境对任务完成质量的影响，探索适应混杂协作式环境的众包模型、收益分配机制以及任务分配机制：1）提出基于群组形成博弈促进多Agent以积极协作的方式执行任务的机制，揭示群组协作在任务完成质量上的优势效应；2）提出基于抽样检测遏制多Agent恶意篡谋行为的机制，激励工作者提供真实有用的任务执行结果；3）提出基于社会准则的混杂非协作多Agent与协作多Agent动态演化众包模型，揭示非协作多Agent向协作多Agent收敛机理。.通过本项目的研究，能够从多Agent博弈的认知角度和建模方法来探索混杂协作式众包机理，有助于大规模群体智能众包系统中任务完成质量优化问题的解决。

本项目主要针对众包环境中工作者非协作与协作混杂行为特征，采用多Agent博弈技术优化任务完成质量。我们首先从多Agent 博弈论角度对工作者非协作性背后的理性动机进行建模。设计抽样检测机制，通过奖励高质量任务完成者并且惩罚低质量任务完成者来激励工作者认真完成任务。然后以优化任务完成质量为目标，设计适应群组博弈的众包模型、收益分配机制以及任务分配机制来建立高质量众包系统。考虑到结构化众包场景，任务发布者处于结构化的社会环境中，任务发布者只能请求到局部工作者，设计图结构约束的激励机制。最后在协作性群体协同场景下，以共享出行，外卖订单服务，安全交通巡逻为典型应用案例的按需移动供应应用，分别设计基于数据驱动的离线群体协同规划和离线策略重利用的在线策略搜索方法，群体多智能体规划方法已申请专利。本项目共发表论SCI、EI论文9篇，其中国际会议3篇（包括国际人工智能顶级会议AAAI 1篇（CCF A类）和多智能体系统顶级会议AAMAS 2篇）；国际期刊6篇（包括IEEE Transactions系列论文4篇等）；国家发明专利1项；研制了一套多智能体规划原型系统。 以该项目为支撑，培养硕士研究生5名，博士研究生3名。截止目前，项目完成预期任务，达到结题指标。

内容获取失败，请点击重试