| When Social Network helps people build various social networking applications, a large number of user information and sensitive data will be collected in the mean time, and through the analysis of these data, some potential privacy information may be disclosed. At present, differential privacy protection model provides a rigorous and quantitative representation of the risk of privacy disclosure, which greatly guarantees the availability of data. In this paper, a generation model of social network publication graph is designed to meet the differential privacy protection. First the social network structure is represented as a graph model, and the original graph is classified into multiple sub graphs according to the characteristics of nodes. Then intensive regional of every sub graph is divided with a Quadtree method, noises of differential privacy protection are added into leaf nodes of the trees, and publication graph is generated by the way of sub graph reconstruction,. Finally, the feasibility and usefulness of the model is verified by the statistical analysis, such as the degree distribution, the shortest path and the clustering coefficient.