Entity Alignment via Neuro-symbolic Reasoning

NeuSymEA is an entity alignment framework driven by neuro-symbolic reasoning, offering an efficient and interpretable solution for large-scale knowledge fusion. This work has been accepted by NeurIPS'25.

🌟 Key Features

• ❶ Low resource requirement: achieving 73.7% hit@1 accuracy on $DBP15K_{FR-EN}$ with only 1% pairs as seed alignments.
• ❷ Explanation over KGs: NeuSymEA's rule mining interpretable alignment results, enabling users to trace the reasoning process of alignment.
• ❸ Scalabile to large KGs: NeuSymEA can handle large-scale KGs efficiently, consistently outperforming state-of-the-art methods.

Scalability analysis on large scale KGs. (Left) Hit@1 alignment performance on large KGs. (Right) Per-second processed entities of neural and symbolic components on different scales of KGs.

🚀 Quick Start

1. Install the required packages

pip install -r requirements.txt

2. Run the baseline model (e.g. LightEA)

python run-baseline.py --dataset fr_en --train_ratio 0.01 --ea_model lightea --gpu 1 

3. Run neuSymEA with the LightEA model as base EA model

python run-neusymea.py --dataset fr_en --train_ratio 0.01 --base_ea_model lightea --gpu 1

4. Run Paris

python run-paris.py --dataset fr_en --train_ratio 0.01

5. Generate interpretations for the inferred pairs using the explainer.

python explain.py

Instances of the mined supporting rules for the inferred pairs in $DBP15K_{FR-EN}$, anchor pairs are shown in bold.

🍀 Citation

If you find this work helpful, please cite our paper:

@inproceedings{chen2025neusymea,
title={NeuSym{EA}: Neuro-symbolic Entity Alignment via Variational Inference},
author={Shengyuan Chen, Zheng Yuan, Qinggang Zhang, Wen Hua, Jiannong Cao, Xiao Huang},
booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
year={2025},
url={https://openreview.net/forum?id=SAbQLqf8XL}
}

Acknowledgement

The code is based on PRASE, Dual-AMN, and LightEA, the dataset is from OpenEA benchmark.

This project is licensed under the GNU General Public License v3.0 (LICENSE).