Italian Regional Cuisine GNN — Evaluation Results
This repository preserves compact evaluation and interpretation artifacts from Leonardo Pacciani-Mori's Italian regional-cuisine graph study. It does not distribute trained checkpoints or source recipe records.
Companion code and full provenance notes:
Portfolio project:
Recorded results
| Approach | Accuracy | Macro F1 | Weighted F1 |
|---|
| Four macro-regions | 59.49% | 52.98% | 58.79% |
| Twenty individual regions | 20.26% | 18.50% | 21.45% |
| Hierarchical classifier | 22.31% | 17.82% | 21.59% |
The committed JSON files contain the complete recorded metrics. signature_ingredients_val.json contains exploratory ingredient-gradient results from the validation split.
Model represented by these results
The classifier used a heterogeneous GAT over recipe, ingredient, and step nodes. Its registered relations connected recipes to ingredients and steps, and steps to ingredients. The richer Neo4j representation also contained tools and sequential step relations, but those were not consumed by the recorded GAT.
The signature analysis used raw ingredient-embedding gradient norms. It was not attention extraction, integrated gradients, or a causal explanation method.
Intended use
These files are retained as transparent portfolio evidence and may support methodological review or comparison. They are not a deployable model release.
Limitations
- No weights or inference package are supplied.
- Source recipe records and train/validation/test splits are not distributed.
- Fine-grained and hierarchical performance is weak and affected by class imbalance.
- Results are from a historical single run and have not been independently replicated from the public checkout.
- Gradient saliency can be unstable and should be interpreted qualitatively.
Licence
The original aggregate result and interpretation files are licensed under
CC BY 4.0. See
LICENSE.md.
Attribution: “Italian Cuisine Graph Analysis, Leonardo Pacciani-Mori.”
No rights to omitted recipe corpora are granted.
Citation
1@misc{pacciani_mori_italian_cuisine_graph,
2 author = {Leonardo Pacciani-Mori},
3 title = {Italian Cuisine Graph Analysis and Regional Classification},
4 year = {2025},
5 url = {https://github.com/LeonardoPaccianiMori/portfolio-italian-cuisine}
6}
