Abstract: To address the problem of information overload caused by the massive manufacturing service information on cloud manufacturing service platforms, a graph neural network-based recommendation method for cloud manufacturing services is proposed in this paper, which effectively overcomes the limitations of traditional recommendation methods that cannot use high-dimensional features of data. Firstly, the features of manufacturing service resources on a platform are extracted, and manufacturing service resources are constructed as a network graph according to different similarity calculation methods. Secondly, a graph sample and aggregate (GraphSAGE) neural network is used for network representation learning, and the learned network features are brought into the link prediction function for model training. Finally, by predicting the link probability among resource nodes, the manufacturing service recommendation for users is completed. Experimental results show that the performance of the link prediction model based on GraphSAGE is better than that of link prediction models based on common neighbors (CN), Adamic-adar (AA) and resource allocation (RA). Thus, better recommendation results are achieved. It provides a theoretical basis for solving the recommendation problem of cloud manufacturing services and helps to improve the decision-making efficiency of users.

Key words: cloud manufacturing, graph neural network, link prediction, manufacturing service recommendation