Query routing is a fundamental problem in unstructured Peer-to-Peer systems. Recently, researches in this area have focused on methods based on query-oriented routing indices. These methods use the historical information of past queries and query hits to build a local knowledge base per peer, which represents the user's interests or profile. Existing approaches represent the user's profile only by some statistics about past queries and they have not addressed two difficult challenging problems: (i) the bootstraping (ii) the unsuccessful relevant peers search. Indeed, when a peer selects an insufficient number of relevant peers from its local knowledge base, it foods the query through the network, which badly affects the routing efficiency and effectiveness. To tackle these problems, we introduce a novel Learning Routing Scheme (LRS). We implemented the proposed scheme and compared its routing efficiency and retrieval effectiveness with a broadcasting scheme (without learning) and a learning scheme taken from the literature. Experimental results show that our scheme carries out better than other ones with respect to accuracy.