Code and datasets for paper "Negative Sampling with Adaptive Denoising Mixup for Knowledge Graph Embedding" accepted by ISWC'23.
We check the reproducibility under this environment.
- Python 3.9.0
- CUDA 10.1
- pytorch-lightning 1.6.5
To run the codes, you need to install the requirements:
git clone https://github.com/DeMix2023/Demix.git
cd Demix
conda create -n demix python=3.9
conda activate demix
pip install -r requirements.txt
You can try our code easily by runing the scripts in ./script, for example:
bash ./script/run_transe_fb.sh
The training process, validation results, and final test results will be printed and saved in the corresponding log file. After training, you can find training logs in ./wandb
.
We put the trained model state dicts in ./output
.
The repository benefits greatly from NeuralKG. Thanks a lot for their excellent work.
Please cite our paper if you use our model in your work:
@inproceedings{Demix,
title = {Negative Sampling with Adaptive Denoising Mixup for Knowledge Graph Embedding},
author = {Chen, Xiangnan and Zhang, Wen and Yao, Zhen and Chen, Mingyang and Tang, Siliang},
booktitle = {{ISWC}},
series = {Lecture Notes in Computer Science},
pages = {253--270},
publisher = {Springer},
year = {2023}
}