Synthesizing agent interactions through the concept of conversation

Abstract

In MultiAgent Based Simulation (MABS), the observation of simulation results is a complex task: interactions between agents produce a large mass of results, which is particularly complex to analyze. We focus on agents' interactions by message exchanges, from the point of view of human observers. In order to facilitate the observation and the analysis of interactions, we propose an upper level of abstraction, compared to the level that focuses on messages, by defining the concept of conversation. A conversation synthesizes information from a set of messages. A conversation is composed of three parts: the metadata, which are synthetic and objective information on the conversation; knowledge, which is specific information that depend on the human observer; and the identifiers of messages that compose the conversation. We illustrate the conversation level with the simulation of animal waste management at a territory scale.