% Encoding: UTF-8
@COMMENT{BibTeX export based on data in FAU CRIS: https://cris.fau.de/}
@COMMENT{For any questions please write to cris-support@fau.de}
@inproceedings{faucris.261480258,
abstract = {A current trend in predictive business process monitoring is to construct predictive models using deep neural networks (DNNs), especially long
short-term memory neural networks, convolutional neural networks, or multilayer perceptron neural networks. While these DNN types typically require data
defined on the Euclidean space (e.g., grids), graph neural networks (GNNs), a relatively new type of DNNs, can compute data defined on the non-Euclidean space
(e.g., graphs). Because GNNs can directly compute graph-oriented data inputs,
generally structured into nodes and edges, they can explicitly model event relationships. This paper investigates gated graph sequence neural networks (GGNNs)
for the next activity prediction. First results with two real-life event logs show that
GGNNs can outperform traditional DNNs regarding predictive quality, especially
if nodes of an input graph are assumed as events, and the graphâ€™s adjacency matrix only describes event relationships of a process instance prefix.