Multi-party communication and information-need anticipation by experience

Abstract

Collaborative computing relies on the modeling and exploiting of team intelligence. While the notion of shared mental models has been widely adopted to explain coordination behaviors in human teams, it becomes increasingly important to investigate the computerization of shared mental models and their application in multi-agent systems. A key element of research along this line is to explore effective ways to developing shared mental models. In this paper, we give a representation model for conversation patterns involving multiple conversation roles. Then, within the R-CAST agent architecture, we detail an approach where agents in a group, via multi-party communication, can anticipate others' information needs using experience-based conversation pattern recognition. The approach can be employed to develop shared mental models among a group for supporting unplanned collaborations.