Abstract: The resource location of unstructured P2P network is usually with low efficiency, and whether a routing message walks in the right direction is not assured. Therefore it is hard to achieve low latency and high query hit rate with low cost under no other supporting mechanism. In this paper, we present a self-organizing semantic cluster based P2P query routing algorithm SOSC. SOSC first categorizes resources shared by each node, and then expresses the semantic of each category by a keywords frequency vector. All of a nodes categories together express its shared resource semantic. Each node tries to establish links with nodes which have most similar category semantic, and therefore SOSC tries to make nodes clustered according to their shared resources semantic. What is more, SOSC transmits a nodes semantic in P2P network by exponentially decaying its keywords frequency vectors of all of its categories. By this way, SOSC creatively solves the problem of cluster semantic expressing and transmitting in a totally distributed environment. SOSC makes a node feel the semantic hierarchy of the semantic of its nearby nodes. Analysis and experiment results show that SOSC is able to achieve high query hit with small routing latency and query cost.