Toward Peer-to-peer Based Semantic Search Engines: an Organizational Approach

This book frames a peer-to-peer information retrievalproblem as a multi-agent framework and attacks itfrom an organizational perspective by exploringvarious adaptive, self-organizing topologicalorganizations, designing appropriatecoordination strategies, and exploiting learningtechniques to create more accurate routing policy forlarge-scale agent organizations. In addition, areinforcement-learning based approach is developed inthis thesis to take advantage of the run-timecharacteristics of P2P IR systems, includingenvironmental parameters, bandwidth usage, andhistorical information about past search sessions. Inthe learning process, agents refine their contentrouting policies by constructing relatively accuraterouting tables based on a Q-learning algorithm. Experimental results show that this learningalgorithm considerably improves the performance ofdistributed search sessions in P2P IR systems. The book is addressed to researchers andpractitioners in information retrieval and searchengine, content-based routing areas.