基于图神经网络的云制造服务推荐方法研究

A Recommendation Method for Cloud Manufacturing Services Based on Graph Neural Networks

针对云制造服务平台上的海量制造服务信息所带来的信息过载问题,提出一种基于图神经网络的云制造服务推荐方法,有效克服了传统推荐方法无法利用数据高维特征的局限性。提取平台上制造服务资源的特征,根据不同的相似度计算方法将制造服务资源构建为网络图;利用邻居采样图神经网络(graph sample and aggregate, GraphSAGE)进行网络的表示学习,并将学习到的网络特征带入链接预测函数进行模型训练;通过对资源节点间的链接概率进行预测,完成对用户的制造服务推荐。结果表明,基于图神经网络算法的链接预测模型,其预测性能要优于所对比的共同邻居(common neighbors, CN)、Adamic-adar (AA)与资源分配(resource allocation, RA)链接预测算法,从而取得较好的推荐效果,为解决云制造服务推荐问题提供理论依据,有助于提高用户的决策效率。

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.