cc_node2vec

This repository evaluates discovered Petri net models against event logs using graph construction, random walks, Word2Vec embeddings, Procrustes alignment, and weighted distance scoring.

Required Inputs

Reproducing the experiments requires both:

• event logs in .xes format
• discovered process models in .pnml format

The .pnml files used in our experiments are included in this repository under data/<dataset>/.

The .xes files are not fully bundled here because of repository size limits. You must place one .xes log inside each dataset folder, for example:

data/
  01-Sepsis/
    data.xes
    data_ilp_reduced.pnml
    data_inductive_reduced.pnml
    data_heuristics_reduced.pnml
    data_split_reduced.pnml
    data_gcn_*.pnml

The benchmark repository of Sommers et al. can be used as a convenient download source, but it is not a strict dependency of this framework. If that source is unavailable, the event logs can still be obtained from public benchmark repositories such as 4TU, and process models can be rediscovered with standard tools such as PM4Py.

Setup

Use Python 3 and install the required packages:

pip install pm4py gensim networkx numpy pandas scipy matplotlib scikit-learn

Run

Main entry point:

python src/main.py --alpha 1.0 --dim 128 --min_count 200 --dataset all --method all --scenario both --lambda_val inf --output_dir results/ --disable_pca_plots

Useful options:

• --dataset all or a single dataset such as sepsis
• --method all or one of ilp, inductive, heuristics, split, gnn
• --scenario static, context_aware, or both
• --disable_pca_plots for batch runs

Batch Runs

Baseline run:

bash ./run_baseline.sh

Ablation run:

bash ./run_ablation.sh

On Windows PowerShell, you can also run the main command directly instead of using bash.

Output

Main results are written to:

results/results_master.csv

Summary tables can be generated with:

python summarize_results.py --input_csv results/results_master.csv --output_dir results/

Reference

Sommers, V. Menkovski, and D. Fahland, "Supervised learning of process discovery techniques using graph neural networks," Information Systems, vol. 115, 2023.