Trust and social networking of multiagent peers

在这项研究中，我们探索了在利用其他代理提供的建议时，在多代理系统中建立信任模型的四个基本挑战。多代理系统是由共存的智能软件代理组成的系统，旨在代表其人类用户解决问题。第一种是能够识别和解决代理人之间的串通行为(其中代理人集体旨在提供欺骗性信息)，例如用于电子商务应用。第二是为界定咨询代理的社交网络设定条件，以便与适当数量的同行以及与能够提供有价值信息的可能性更大的同行合作。第三是将代理信任模型投射到两个非常热门但关键的应用程序中，这些应用程序面临动态变化的挑战：交通(车辆代理共享信息，以帮助管理城市交通)和医疗保健(利用受信任同行的社交网络提供信息，以教育用户了解当前的医疗问题)。最后一个项目将更仔细地研究特定于上下文的信任建模，表明多代理系统的组成(利他代理或恶意代理)和用户的倾向(风险厌恶或风险寻求)将对其他代理的信任建模设置不同的要求。总之，本研究旨在通过对共存的其他代理进行推理的新技术，使智能代理能够在多代理环境中改进决策。随着我们解决这些关键场景的扩展，将出现一种具有重要适用性的信任建模方法。此外，对有效界定社交网络的赞赏将会出现。

In this research, we explore four fundamental challenges to modeling trust in multiagent systems -- systems of co-existing intelligent software agents that are designed to problem solve on behalf of their human users -- when making use of advice provided by other agents. The first is to be able to identify and address collusive behaviour among agents (where groups of agents collectively aim to provide deceptive information) for example for e-commerce applications. The second is to set the conditions for delimiting the social network of advising agents, in order to operate with an appropriate number of peers and with peers that have a greater likelihood of being able to provide information of value. The third is to project models of agent trust into two very topical but critical applications with challenges of dynamic change: that of transportation (vehicular agents sharing information to assist in the management of urban traffic) and healthcare (leveraging a social network of trusted peers to provide information to educate a user about a current medical problem). The final project will look more closely at context-specific trust modeling, suggesting that the makeup of the multiagent system (altruistic or malicious agents) and the proclivity of the user (risk averse or risk seeking) will set differing demands for the trust modeling of other agents. In all, this research aims to enable improved decision making by intelligent agents in multiagent environments through novel techniques for reasoning about the other agents that co-exist. With our extensions to address these critical scenarios what will emerge is a methodology for trust modeling that has a significant reach of applicability. In addition, an appreciation for effective delimiting of social networks will emerge.