Routing algorithms in delay tolerant networks (DTN) adopt the storecarry-forward way, and this needs the nodes to work in a cooperative way. However, nodes may not be willing to help others in many applications and this behavior can be seen as individual selfish. On the other hand, nodes often can be divided into different communities, and nodes in the same community often have some social ties. Due to these social ties, nodes are more willing to help the one in the same community. This behavior can be seen as social selfish. Note that some nodes may belong to more than one community in the real world, and this phenomenon makes the network have overlapping communities. This paper proposed a theoretical model to describe the performance of epidemic routing (ER) in such network. Simulation results show the accuracy of our model. Numerical results show that the selfish nature can make the performance of the routing policy be worse, but those nodes belonging to multi-communities can decrease the impact of the selfish nature in certain degree.