GraphMatcher

GraphMatcher is a graph-based ontology matching system that aligns entities from two ontologies using a neural architecture centered on graph attention and neighborhood-aware representations.

The project combines ontology preprocessing, graph construction, and a graph attention network to predict likely alignments between concepts and properties across two RDF/OWL sources.

Citation

If you use GraphMatcher in your work, please cite:

@inproceedings{efeoglu2022graphmatcher,
  title = {GraphMatcher: A Graph Representation Learning Approach for Ontology Matching},
  author = {Efeoglu, Sefika},
  booktitle = {Proceedings of the 17th International Workshop on Ontology Matching (OM 2022)},
  year = {2022},
  series = {CEUR Workshop Proceedings},
  note = {Co-located with ISWC 2022}
}

Highlights

• Ontology alignment for RDF/OWL-based datasets
• Graph attention network for neighborhood-aware matching
• Support for entity and property matching
• Config-driven dataset and model paths
• Python 3.9-compatible dependency set

Project goals

This project follows the idea that ontology alignment can benefit from structural context, not just lexical similarity. In practice, the model learns from neighboring concepts and relations to estimate whether two entities represent the same concept across ontologies.

The work was developed for the OAEI conference track and was designed to score strong matches in uncertain reference alignment settings.

Repository structure

.
├── config.ini
├── requirements.txt
├── datasets/
│   └── conference/
│       ├── ontologies/
│       └── alignments/
├── outputs/
├── saved_models/
├── src/
│   ├── model/
│   ├── preprocessing/
│   ├── train_model.py
│   ├── test_model.py
│   └── project_paths.py
├── tests/
│   └── test_project_paths.py
└── README.md

Setup

1. Create a virtual environment and install dependencies:

python3.9 -m venv .venv39
source .venv39/bin/activate
python -m pip install --upgrade pip
python -m pip install -r requirements.txt

2. Configure the project in config.ini:

[General]
dataset = conference
K = 5
ontology_split = False
max_false_examples = 150000

[Paths]
dataset_folder = datasets
alignment_folder = /alignments/
save_model_path = saved_models/conference.pt
load_model_path = saved_models/conference.pt
output_folder = outputs/

[Parameters]
max_paths = 21
max_pathlen = 8

[Hyperparameters]
lr = 0.001
num_epochs = 5
weight_decay = 0.001
batch_size = 32

The project resolves paths relative to the repository root, so local absolute paths are no longer required.

Training

Run the training pipeline from the project root:

python src/train_model.py

Testing

Evaluate alignment between two ontology files:

python src/test_model.py path/to/source.owl path/to/target.owl

Example alignment output

<map>
  <Cell>
    <entity1 rdf:resource='http://conference#has_the_last_name'/>
    <entity2 rdf:resource='http://confof#hasSurname'/>
    <relation>=</relation>
    <measure rdf:datatype='http://www.w3.org/2001/XMLSchema#float'>0.972</measure>
  </Cell>
</map>

Notes

• This codebase uses rdflib for ontology parsing instead of the legacy ontospy package.
• The project is validated for Python 3.9, which is the recommended runtime for the pinned dependency set.
• The path logic was centralized in src/project_paths.py so the project behaves consistently across machines.

References

[1] Iyer, Vivek, Arvind Agarwal, and Harshit Kumar. "VeeAlign: Multifaceted Context Representation Using Dual Attention for Ontology Alignment." Proceedings of EMNLP 2021.

[2] Veličković, Petar, et al. "Graph Attention Networks." arXiv preprint arXiv:1710.10903 (2017).

Acknowledgements

This project builds on the VeeAlign design and uses a graph attention approach inspired by the GAT paper above.