Triple Feature Propagation (TFP)

This is the code of Triple Feature Propagation (TFP) introduced in our paper: "Rethinking Smoothness for Fast and Adaptable Entity Alignment Decoding", accepted by NAACL 2025 as Findings. It is the formal version of the paper "Gradient Flow of Energy: A General and Efficient Approach for Entity Alignment Decoding".arxiv

Datasets

The dataset and the embedding we processed can be downloaded at GoogleDrive. You can also use the same datasets in DATTI.

• ent_ids_1: ids for entities in source KG;
• ent_ids_2: ids for entities in target KG;
• rel_ids_1: ids for relations in source KG;
• rel_ids_2: ids for relations in target KG;
• sup_ent_ids: training entity pairs;
• ref_ent_ids: testing entity pairs;
• triples_1: relation triples encoded by ids in source KG;
• triples_2: relation triples encoded by ids in target KG;

Just run main.py

Environment

• Python == 3.7.0
• tensorflow == 2.6.0
• Numpy
• tqdm

Acknoledgement

We appreciate DATTI for their open-source contributions.

Cite

Please consider citing this paper if you use the code or data from our work. Thanks a lot ~

@article{wang2024gradient,
  title={Gradient Flow of Energy: A General and Efficient Approach for Entity Alignment Decoding},
  author={Wang, Yuanyi and Sun, Haifeng and Wang, Jingyu and Qi, Qi and Sun, Shaoling and Liao, Jianxin},
  journal={arXiv preprint arXiv:2401.12798},
  year={2024}
}