Due to the openness and anonymity of PeertoPeer (P2P), the types of interactions between nodes are complex and diverse, and they have small repetitive interactions, which makes it difficult to define the trust relationship between nodes. Based on selected nodes, we propose a P2P network trust model, and obtain comprehensive trust value through direct trust and the recommended trust for each node. Recommendation nodes whose trust value is closest to the interactive nodes are selected, that is to say, nodes with the smallest trust difference are selected. We also propose a joint punishment mechanism and an incentive mechanism to prevent gang cheating of the conspiracy group. Simulation experiments show that the trust model is not only capable of resisting attacks of malicious nodes, but also has a high selfadaptability.