Deep Reinforcement Learning for Entity Alignment

This repository is the official implementation of Deep Reinforcement Learning for Entity Alignment, Findings of ACL 2022.

Requirements

Please first download the dataset from OpenEA, and then install gym and required packages of OpenEA:

conda create -n openea python=3.6
conda activate openea
conda install tensorflow-gpu==1.8
conda install -c conda-forge graph-tool==2.29
conda install -c conda-forge python-igraph
pip install -e .
pip install gym

Training & Evaluation

Use the following scripts to run RLEA with RDGCN as basic EEA model on D-Y:

cd run
python runRLEA.py --model_name rdgcn --dataset D_Y

If run with the stored embeddings:

python runRLEA.py --model_name rdgcn --dataset D_Y --restore_embeddings True

For SEA which has projection matrices:

python runRLEA.py --model_name sea --dataset D_Y --mapping True

Available EEA models: JAPE, SEA, RSN, RDGCN, AlignE, BootEA.

The corresponding model names: jape, sea, rsn, rdgcn, aligne, bootea.

Available datasets: EN_FR, EN_DE, D_W, D_Y

Results

Our model achieves the following performance on OpenEA V1 benchmarks:

How to Cite

If you use or extend our work, please cite the following paper:

@inproceedings{RLEA,
    title = "Deep Reinforcement Learning for Entity Alignment",
    author ="Lingbing Guo and
            Yuqiang Han and
            Qiang Zhang and
            Huajun Chen",
    booktitle = "Findings of ACL 2022",
    url = "https://openreview.net/pdf?id=CRBzhRdkycU",
}

Contributing

This project is licensed under the MIT License. Any contribution is welcomed and appreciated.