Identifying Key Information on Life Cycle of Engineering Data by Graph Convolutional Networks and Data Mining

Abstract

Engineering data, including product data-conversion networks and software dependency networks, are very important for the long-term preservation of product information. With the explosive growth of data in recent years, product information has become increasingly diversified and complex, which poses new challenges to the long-term preservation of product data. A better understanding of the functions of complex networks can help us take more effective measures to maintain and control such complex systems, and then adopt more effective methods to achieve life cycle management. It is currently difficult for traditional heuristic methods to deal with such large-scale complex systems. In recent years, however, the use of graph neural networks to identify key nodes attracted widespread attention, but this requires a large amount of training data. It is difficult to obtain large-scale relational data and establish identification models in engineering fields. Combining a graph convolution network with a data-mining method, a key node identification method in a graph convolution network based on data mining is proposed. The method first determines the type of complex network according to the power-law distribution and centrality of the network and then uses the corresponding evolutionary model to generate a large-scale synthetic network to effectively train the model. The experimental results from two real networks show that this method improves the identification performance of key nodes by using synthetic data with the same characteristics as the real network, and provides a new perspective for product life cycle management.