The key to the usability of large-scale decentralize peer-to-peer (P2P) systems, and one of the most challenge design aspects, is efficient mechanism for distributed resource discovery. Unstructured P2P networks are very attractive because they do not suffer the limitations of centralized systems an the drawbacks of highly structured approaches. However the search algorithms are usually based on simple flooding scheme generating large loads on the network participants. In this paper to address this major limitation, we present the design an evaluation of an innovative searching protocol in unstructured P2P networks. The approach aims at dynamically adapting the network topology to peers' interests, on the basis of a peer neighbor selection algorithm. Each peer builds and maintains profiles of other peers, describing their interests and resources. Given a query, it is consequently routed according to the predicted match with other peers' profiles. Experimental evaluation shows that the approach is able to exploit query interactions among users, in order to dynamically group peer nodes in clusters containing peers with shared interests and organized into a small world topology.