Home

Awesome

KECG

Source code and datasets for EMNLP 2019 paper "Semi-supervised Entity Alignment via Joint Knowledge Embedding Model and Cross-graph Model".

Reqirements

Code

This implementation includes KECG, KECG(w\o K), and KECG(w\o NNS). Example of running KECG on DBP15K(ZH-EN)

CUDA_VISIBLE_DEVICES=0 python3.6 run.py --file_dir data/DBP15K/zh_en --rate 0.3 --lr 0.005 --epochs 1000

Example of running KECG(w\o K)

CUDA_VISIBLE_DEVICES=0 python3.6 run.py --file_dir data/DBP15K/zh_en --rate 0.3 --lr 0.001 --epochs 500 --wo_K

Example of running KECG(w\o NNS)

CUDA_VISIBLE_DEVICES=0 python3.6 run.py --file_dir data/DBP15K/zh_en --rate 0.3 --lr 0.05 --epochs 1000 --wo_NNS

Dataset

The used datasets DBP15K and DWY100K are from subfolder named "mapping" of BootEA and JAPE. (But need to combine "ref_ent_ids" and "sup_ent_ids" into a single file named "ill_ent_ids" before running KECG.) Here, you can directly unpack the data file

unzip data.zip

Acknowledgement

We refer to some codes of these repos: pyGAT, BootEA, GCN-Align. Appreciate for their great contributions!

Cite

If you use the code, please cite this paper: