An Empirical Study of Selective Overhearing in Hierarchical Organizations∗

Overhearing is a plan recognition approach for monitoring multi-agent systems, by listening to the routine inter-agent communications. Previous investigations of overhearing explored an extensive set of techniques for overhearing, mostly relying on the assumption that all inter-agent communications are accessible to the overhearing agent. However, in real-world settings, overhearing resources are limited, and thus the overhearing agent must be selective in carefully choosing the conversations it will overhear. This paper presents an empirical study of such selective overhearing. We focus on overhearing hierarchical organizations that are common in the real-world settings. We first present a model of the conversations expected in such organizations. We then present the results of extensive experiments with several overhearing strategies, particularly suited for such organizations. Based on these extensive experiments, we are able to isolate the parameters influencing their behavior. We reach several qualitative conclusions.