Process Mining is increasingly shifting to graph-based representations of event data, enabling the application of Graph Neural Networks (GNNs) for various process mining tasks.
Recent Master projects developed the first GNN-based process discovery algorithm which trains a GNN to translate an event log into a process model. The GNN achieves state-of-the-art performance in terms of accuracy and simplicity.
- https://research.tue.nl/en/studentTheses/process-discovery-using-machine-learning
- https://research.tue.nl/en/studentTheses/applying-transfer-learning-to-gnn-based-automated-process-discove
While this demonstrates the potential of GGNs, training of GNNs for process discovery is not robust (results are difficult to replicate) and many other potential applications of GNNs for process mining have not been explored yet.
This Master project aims to explore and improve applications of GNNs to process mining. Various directions are possible
- Improving robustness and efficiency of GNN-based process discovery techniques
- Extending capabilities of GNN-based process discovery for model evolution and repair
- Extending capabilities of GNN-based process discovery towards object-centric processes (considering more complex input graph structures)
- GNNs for automated outlier analysis in object-centric event data (training GNNs to learn distributions of complex graph structures in object-centric event data and then detecting outliers)
- Other ideas and suggestions of students are welcome
Contact: Dirk Fahland (d.fahland@tue.nl)