A hybrid graph neural network-based federated learning method for personalized manufacturing service composition recommendation

The demand for personalized manufacturing service recommendations is expanding with the popularity and application of industrial Internet platforms. However, the recommendation system has drawbacks in data privacy and security when exchanging parameters of clients. Therefore, this paper proposes a hybrid graph neural network-based federated learning method for personalized manufacturing service composition recommendation (FLGRC). First, a hybrid differential privacy algorithm based on federated learning is designed to solve the data island problem and achieve collaborative training. Second, an improved method of data mining is established to discover the collaborative relationships between different enterprises. Third, the graph neural network algorithm is employed to predict missing QoS (Quality of Service) data, and the lists of recommendations are generated in accordance with fast non-dominated sorting and Top-N sorting rules. Finally, a real industry Internet platform case is adopted in this paper. The experiments analyze the accuracy of the prediction results. Moreover, the results obtained from the proposed algorithm are compared with those collected from other recommendation algorithms to verify the recommendation effect of the model.