KROMA
Ontology matching with knowledge retrieval and large language models
Documentation · Quick start · Example · Results · Paper · Citation
KROMA finds the concepts that two ontologies share. It retrieves likely matches for each concept, accepts identical names directly, and asks a large language model about the rest, showing it each concept's labels, synonyms, parents and children.
Quick start
pip install kroma-om openai
export OPENAI_API_KEY=sk-...
kroma match mouse.owl human.owl -o alignments.csv
Or from Python:
import kroma
result = kroma.match("mouse.owl", "human.owl", model="gpt-4o-mini")
result.save("alignments.csv") # .csv, .tsv, or .rdf (OAEI format)
KROMA reads any RDF format (.owl, .ttl, .rdf, …) and works with OpenAI,
Together AI, or any function that takes a prompt and returns text. See the
documentation for all
options.
Example
Try KROMA with a local Ollama model, no API key needed:
git clone https://github.com/cwru-sdle/kroma.git && cd kroma/examples
pip install kroma-om
ollama pull llama3.2:1b
python match_ollama.py --model llama3.2:1b
| Source | Target | Confidence (0-10) |
|---|---|---|
| heart | cardiac organ | 10 |
| lung | lungs | 8 |
| organ | body organ | 8 |
Run with llama3.2:1b on a small demo dataset; see the
example for details.
Results
F1 (%) on six OAEI tracks, from the ISWC 2025 paper:
| Track | KROMA | Best baseline |
|---|---|---|
| Mouse–Human | 94.94 | 92.20 |
| NCIT–DOID | 98.63 | 94.80 |
| Nell–DBpedia | 97.08 | 96.00 |
| YAGO–Wikidata | 95.54 | 92.19 |
| ENVO–SWEET | 93.18 | 83.70 |
| MI–MatOnto | 61.25 | 32.97 |
Ablations, other models and cost are in the full paper.
Reproducing the paper
git clone https://github.com/cwru-sdle/kroma.git && cd kroma
./setup.sh && source .venv/bin/activate
uv pip install ".[models]"
kroma run --method_config kroma_scibert_envo_sweet \
--llm meta-llama/Llama-3.3-70B-Instruct-Turbo-Free
Datasets, configs and flags are described in Benchmark experiments.
Citation
@inproceedings{nguyen_2025_kroma,
title = {KROMA: Ontology Matching with Knowledge Retrieval and Large Language Models},
author = {Lam Nguyen and Erika Barcelos and Roger French and Yinghui Wu},
booktitle = {The Semantic Web -- ISWC 2025: 24th International Semantic Web Conference, Proceedings, Part I},
series = {Lecture Notes in Computer Science},
publisher = {Springer},
pages = {629--649},
year = {2025},
doi = {10.1007/978-3-032-09527-5_34}
}
License
BSD 3-Clause · Developed at the SDLE Research Center, Case Western Reserve University.
Metadata
Release files for kroma-om 0.2.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| kroma_om-0.2.0.tar.gz | 57.5 kB | With provenance | Details for kroma_om-0.2.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| kroma_om-0.2.0-py3-none-any.whl | Python 3 | none | any | Details for kroma_om-0.2.0-py3-none-any.whl |
Total release size: 121.1 kB
Release files / kroma_om-0.2.0.tar.gz
| Download URL | kroma_om-0.2.0.tar.gz |
|---|---|
| Size | 57.5 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
8ccf0ffba9ae07e8ecdb3393c04754ecca27465bbe9efa47a9e93622c87e4f6a
|
|
BLAKE2b-256 checksum How to use checksums |
01db449669d902a98e0608c7e71635b7bd0c80be707aa9f88ad0e25fb16c265c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 4, 2026.
Transparency logRelease files / kroma_om-0.2.0-py3-none-any.whl
| Download URL | kroma_om-0.2.0-py3-none-any.whl |
|---|---|
| Size | 63.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
ae6e36dc5c48c29b468594b2ab8b867bdcfe450c617daf7b3a7cc7a7d544c8d6
|
|
BLAKE2b-256 checksum How to use checksums |
4d9db6ff0307db4140200f8ede108bfeac4d78561daba959e8279c5e9d9f865e
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 4, 2026.
Transparency logSecurity for kroma-om 0.2.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
All 2 files in this release have been uploaded with provenance metadata.
Attestation for kroma_om-0.2.0.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 4, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.