In this talk, I will describe research on query routing on unstructured peer-to-peer networks, where each node contains a collection of documents and is able to evaluate a query it received against the collection. An essential operation for query routing is to be able to select a node among many neighbors to route the query so that it is either evaluated at that node or further routed to another node for evaluation. This process is called collection ranking in traditional IR and is essential for federated search and meta-search as well. I will discuss some research results conducted by my research group on this problem, including a method using Markov Decision Process (MDP) to precompute an optimal routing strategy for single-term queries and methods that utilize term correlation to obtain precise routing decision for multi-term queries. Concluding the talk, I will outline a few research directions that we are taking, including the use of clickthrough information captured at a search node for adapting the search results for a user community.
