Multi-agent knowledge integration mechanism using particle swarm optimization

Abstract

Unstructured group decision-making is burdened with several central difficulties: unifying the knowledge of multiple experts in an unbiased manner and computational inefficiencies. In addition, a proper means of storing such unified knowledge for later use has not yet been established. Storage difficulties stem from of the integration of the logic underlying multiple experts' decision-making processes and the structured quantification of the impact of each opinion on the final product. To address these difficulties, this paper proposes a novel approach called the multiple agent-based knowledge integration mechanism (MAKIM), in which a fuzzy cognitive map (FCM) is used as a knowledge representation and storage vehicle. In this approach, we use particle swarm optimization (PSO) to adjust causal relationships and causality coefficients from the perspective of global optimization. Once an optimized FCM is constructed an agent based model (ABM) is applied to the inference of the FCM to solve real world problem. The final aggregate knowledge is stored in FCM form and is used to produce proper inference results for other target problems. To test the validity of our approach, we applied MAKIM to a real-world group decision-making problem, an IT project risk assessment, and found MAKIM to be statistically robust.