The research community has turned to create Semantic Overlay Networks for information retrieval in large scale peer-to-peer networks. Much research work in semantic overlay protocols and searching algorithms is done and their results indicate that semantic overlay is powerful for content searching in peer-to-peer network. However, very limited work has been done in analyzing and evaluating characteristic about semantic overlay network. In this paper we identify a natural property of semantic overlay network, community structure. We setup a mathematical model to evaluate the community structure property. A heuristic backtrace-greedy hybrid algorithm is also designed to optimize the evaluation value of a given overlay network. Using the evaluation model we compare the SemreX semantic overlay with the Gnutella network. We find that the SemreX overlay network has a distinct feature of community structure, while the Gnutella network does not have such feature.