FLORA

This is a Python implementation of the paper FLORA: Unsupervised Knowledge Graph Alignment by Fuzzy Logic (best paper award at ISWC 2025).

FLORA is an unsupervised system for automatic knowledge graph (KG) alignment, jointly matching entities and relations in one KG to their equivalents in another.

FLORA is a simple yet effective method that (1) is unsupervised, i.e., does not require training data, (2) provides a holistic alignment for entities and relations iteratively, (3) is based on fuzzy logic and thus delivers interpretable results, (4) provably converges, (5) allows dangling entities, i.e., entities without a counterpart in the other KG, and (6) achieves state-of-the-art results on major benchmarks.

FLORA extends PARIS system, which had three key limitations: (1) no convergence guarantees, (2) poor performance when functional relations are absent, and (3) the inability to see literal similarities beyond a strict identity.

Running FLORA

Check the preconditions. Similar to PARIS, FLORA needs two knowledge graphs, which each contain: (1) a large number of instances. (2) a limited number of relations, (3) a large number of facts between instances, (4) a large number of facts between an instance and a literal. The input KGs have to be in Turtle format.

Set up the environment. Clone this repository and set up the environment via "requirements.txt". Supports Python >=3.9, <3.12.

pip install -r requirements.txt

Pre-compute the string embeddings. To initialize literal similarities, FLORA needs embeddings for all strings (excluding dates and numbers). To produce these embeddings separately, run:

python literals.py <kg1> <kg2> <embedding_path>

For example:

python literals.py ../data/kg1.ttl ../data/kg2.ttl ../data/emb/

Here, kg1 and kg2 are paths to the knowledge graphs in Turtle format, and embedding_path is a path to a folder where the embeddings can be stored.

Run the Code.

To align two KGs, adapt the following command to your case:

python main.py --kg1 ../data/kg1.ttl --kg2 ../data/kg2.ttl --embedding ../data/emb/ --output results.ttl

If the embdding path is not provided or does not exist, embeddings will be automatically computed before performing the alignment. Optional parameters can be set with --alpha, --init, and --epsilon, see our paper or run python main.py --help for a description. If training data is available, specify its file path with --trainingdata parameter. The training file can be in any format (e.g., .txt, .ttl, .csv) but contain aligned entities between two KGs.

Reproducing the experiments

Dataset. FLORA uses multiple datasets from different sources:

• OpenEA: D_W_15K_V1 and D_W_15K_V2
• DBP15K: fr_en, ja_en, zh_en
• OAEI KG Track: memoryalpha-stexpanded, starwars-swtor

We also provide two mini-test datasets: Person, Restaurant from OAEI 2010 for quick test. For detailed statistics on each dataset, please refer to statistics.pdf.

Due to memory limitations, all datasets and pretrained embeddings used in the paper are on the drive. Download and unzip all files in the data folder.

To produce the alignment results, use the following command for existing datasets:

python main.py --dataset OpenEA/D_W_15K_V2/ --embedding emb/D_W_15K_V2/ --alpha 3.0 --init 0.7 --output dw-v2.ttl

If training data is available, run:

python main.py --dataset OpenEA/D_W_15K_V2/ --embedding emb/D_W_15K_V2/ --trainingdata OpenEA/D_W_15K_V2/721_5fold/1/train_links --alpha 3.0 --init 0.7 --output dw-v2-sup.ttl

You can also run bash run.sh to reproduce the results.

Evaluation and Analysis.

The original alignment results (generated by running main.py) are stored in the save folder by default. To obtain clean results or to evaluate and analyze them, run analysis.ipynb block by block, adjusting gold standard path REF_PATH and results path RES_PATH as necessary.

Citation

If you use this project for academic purposes, please cite our paper:

@inproceedings{FLORA,
    title = "FLORA: Unsupervised Knowledge Graph Alignment by Fuzzy Logic",
    author = "Peng, Yiwen and Bonald, Thomas and Suchanek, Fabian",
    booktitle = "International Semantic Web Conference (ISWC)",
    year = 2025
}