Developing a Method for Building Business Process Models Based on Graph Neural Networks in the Absence of Task Identifier Data
The contemporary methodology of business process modeling is closely tied to process mining. The aim of the study is to develop a method of creating business process models through the restoration of links between events recorded in logs in the absence of CaseID data based on graph neural networks....
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Qubahan
2024-01-01
|
| Series: | Qubahan Academic Journal |
| Online Access: | https://journal.qubahan.com/index.php/qaj/article/view/333 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The contemporary methodology of business process modeling is closely tied to process mining. The aim of the study is to develop a method of creating business process models through the restoration of links between events recorded in logs in the absence of CaseID data based on graph neural networks. The problem is solved by applying the graph convolutional networks architecture. The study employs a combination of a weighted adjacency matrix and an adjacency matric accounting for the graph data structure. Textual information about the tasks involved in the business process is considered when implementing the feature matrix using embeddings. The Navec embedding model is chosen to represent task titles as numerical vectors. The study was based on parsing the technological log of the 1C:Enterprise system. The obtained solutions make it possible to restore the required connection (Sequence flow) in the model of the "Approval of a commercial offer" business process in the absence of Case ID data in the event log as part of the "Reset request" task.
|
|---|---|
| ISSN: | 2709-8206 |