Engineering graphs describe trace links between data of different applications. Creating trace links manually can be time-consuming and error-prone. Based on graph patterns, several automatic approaches can predict missing links (links that are likely to exist):
- Deep learning
- Graph mining
Traditional link prediction approaches are not perfectly suitable for engineering graphs which have complex patterns and are typically small in size compared to large social network graphs. Furthermore, engineering graphs evolve over time. The individual sets of changes to an engineering graph can be used to enhance the link prediction accuracy. Below is an image of the most recent neural network architecture that we used for link prediction using Graph Convolution Embedded LSTM.