There are deficiencies of the interest-based driven P2P search method in discovering interest of node and expanding interest in the context of semantic search. This paper improves the Social-P2P algorithm and puts forward a search method considering the search behavior and the content of nodes. The method introduces the concept lattice theory. According to the node content and user search behavior, a friend list is established, then formal context is etracted from the triend list to build a concept lattice, and the interest domain. It searches for messages within the query concept lattice, cuts down the search path and reduces the search message. Concepts which have deflection order expands the context of semantic query messages and enhance the search accuracy. Experimental results indicate that compared with the traditional flooding method, Social-P2P search method has better recall race and accuracy.