P2P systems are a great solution to the problem of distributing resources. The main issue of P2P networks is that searching and retrieving resources shared by peers is usually expensive and does not take into account similarities among peers. In this paper we present preliminary simulations of PROSA, a novel algorithm for P2P network structuring, inspired by social behaviours. Peers in PROSA self--organise in social groups of similar peers, called ``semantic--groups'', depending on the resources they are sharing. Such a network smoothly evolves to a small--world graph, where queries for resources are efficiently and effectively routed.