In the implementation of project operations, it is necessary to find suitable participants. However, the traditional manual method to find participants is time-consuming and labor-intensive, and a more advanced and effective method needs to be explored. The recommendation system can provide people with a recommendation service for finding participants, which greatly solves the problem of low efficiency and quality in finding participants in project work. This paper constructs a recommendation system for project collaboration participants based on collaborative filtering algorithm(CFA), and builds a collaborative relationship network. Using a graph neural network(GNN) to calculate the edge weights of the network, the deep learning(DP) model is introduced into the recommendation system(RS). It is found that the recommendation accuracy of the CFA has been improved, which verifies that the classical CFA can improve the effectiveness after the introduction of the deep neural graph network. The experimental analysis of the recommendation application of the RS to project participants shows that the more obvious the link relationship of the participants is, the higher the probability that the participant is recommended by the DP model in optimizing the collaborative management algorithm.