.png){kind=icon}
.png){kind=icon}
Return
MAKEE: Multi-view attribute network and sequence embedding approach for predictive process monitoring
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2025.114299.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Predictive business process monitoring can help detect and solve problems on time by monitoring the execution of business processes in real time, thereby improving overall business efficiency and performance. Current deep learning-based studies have found that embedding structural information of process models helps neural networks learn the deep logic behind business processes. However, they mainly focus on the control-flow perspective, while other perspectives behind the business process, such as organizational structure, social network, and resource behavior, have been largely overlooked. To address this issue, this study proposes a multi-view learning prediction approach that integrates complementary information from both multiple attribute networks and sequences. We carefully design a deep learning model framework to integrate multi-view structural and sequential information for the next-activity prediction of the running trace. On the one hand, a simple and efficient process mining algorithm is designed to model multiple attribute network graphs, and a graph convolutional network is integrated to learn their multi-view structural information, helping understand the deep features of business scenarios. For this, a node feature enhancement method is proposed to integrate global information from historical business executions to help the proposed neural network understand the structure of a complete business scenario. On the other hand, we construct the feature representation of attribute sequences and integrate the Transformer to capture the dependency relations and sequential features within attribute sequences. Experimental evaluation of twelve real-life event logs shows that the proposed approach performs well in prediction accuracy and robustness.
Keywords:
business process monitoring
multi-view learning
deep learning
next-activity prediction
graph convolutional network
Journal
K
IF:
7.6
Papers:
1.2W
Citations:
4.5W