Task-based Multi-agent Coordination for Information Gathering
Katia P. Sycara Dajun Zeng
The Robotics Institute
School of Computer Science
Carnegie Mellon University
Pittsburgh, PA 15213
The ubiquity of network-based information resources has given impetus for the development of intelligent software agents that will be able to (1) gather task-related information automatically or with little help from human users from various on-line information resources, (2) resolve potential conflicts among acquired knowledge from different information resources, and (3) more importantly, collectively and effectively solve tasks requested by human users.
We report on our work on developing distributed collections of intelligent information agents that cooperate asynchronously to perform goal-directed information retrieval and information integration in support of various tasks, such as finding information about people on the Internet, managing calendars and making arrangements to host visitors in an academic environment. The task of hosting a visitor involves arranging the visitor's schedule with faculty that match the interests that the visitor has expressed in his/her visit request. The visitor hosting task is one of the tasks that are investigated in the context of the PLEIADES project at Carnegie Mellon University. The broader goal of PLEIADES is to characterize and develop distributed agent-based architectures that are composed of negotiating and learning agents and apply them to tackle information and activity management problems for everyday use. Agents coordinate and negotiate with each other to resolve disparities in the retrieved information. In addition, they learn from their users, the information sources, and each other.
In this paper, we will focus on the architecture of our distributed system, and the interactions among its agents for task-based accessing of heterogeneous, distributed information sources. We will use the visitor hosting task as an example domain for illustration.
2. Architecture for Cooperative Intelligent Information Retrieval
In the collection of agents we have developed, we distinguish two types of agents: taskspecific software agents help users perform tasks by communicating with each other and/or querying and exchanging information with information-specific software agents, which provide intelligent access to a heterogeneous collection of information resources. For example,