Task Allocation of Multiagent Groups in Social Networked Systems

Abstract

In large-scale multiagent systems (MASs), agents are often organized in groups through social networks. The advantages of group effectiveness cannot be fully achieved with traditional algorithms based on task decomposition or team formation. Therefore, a group-oriented method for task allocation is proposed in this article, where factors, such as skills, capabilities, communication costs, reputations, and social context are considered to model the comprehensive social value of groups. When a task appears, candidate workers can be selected in each group through internal negotiation, which will be utilized to calculate the group value for the task with the established value model. Then, the comprehensive social value of the group can be obtained based on the combination of its self-value and the social context value of its neighbor groups. The group with the highest comprehensive social value will then be allocated to the task. When an agent group cannot execute a task independently, it can seek assistants from neighbor groups through the social network. After the completion of a task, agents will obtain a reputation reward based on the quality of task completion and their contributions to the task, which can promote the development of social networks. The experiments show that the proposed social group-oriented method, compared with the team formation and nonsocial methods, can achieve better performance of task allocation in the success rate, effectiveness, efficiency, and collaboration.