A multi-agent approach for peer-to-peer-based information retrieval systems

Abstract

This paper develops and analyzes distributed search techniques for use in a peer-to-peer (P2P) network-based Information Retrieval (IR) system. In the absence of a centralized mediator with global knowledge that directs requests to appropriate agents, agents must cooperate to forward the queries among themselves so as to find appropriate agents, and return and merge the results in order to fulfill the information retrieval task in a distributed environment. In our approach, the agent society is connected through an agent-view structure maintained by each agent. Initially, the agent-view structures are formed by agents connecting to each other randomly. However, we show that such an approach can be significantly enhanced by dynamically reorganizing the underlying agent-view topology and deploying contextsensitive distributed search algorithms. Experimental results indicate that appropriate organizational structures and distributed search mechanisms can have a positive influence on system performance.