We first discuss the semantic heterogeneity problem of interaction in large scale MAS. Based on agent society theory, we introduce a grouping method to decompose MAS by different domain or concept, and propose an ontology mapping mechanism to resolve the heterogeneity problem. In the agent society environment, we discuss both inter-society and intra-society ontology mapping method, and finally introduce the architecture of prototype system for ontology mapping in agent society. We also use user participation to improve our mapping quality. Our research is based on agent society environment, mainly focusing on interoperability between societies and interaction in society. Based on our research and analysis, we intend to do some works for future research activities.