For resource discovery in social networks, people can directly contact some acquaintances that have knowledge about the resources they are looking for. However, in current peer-to-peer networks, peer nodes lack capabilities similar to social networks, making it difficult to route queries efficiently. In this paper, we present a social-like system (Social-P2P) for resource discovery by mimicking human behaviours in social networks. Different from most informed search algorithms, peer nodes learn knowledge from the results of previous searches and no additional overhead is required to obtain extra information from neighbouring nodes. Unlike community-based P2P information sharing systems, we do not intend to create and maintain peer groups or communities consciously. Peer nodes with the same interests will be highly connected to each other spontaneously. Social-P2P has been simulated in a dynamic environment. From the simulation results and analysis, Social-P2P achieved better performance than current methods.