- Python 3.6+
- PyTorch 1.0+
- NumPy 1.17.2+
To reproduce the results of RQE and HRQE on WN18RR, FB15k237, WN18 and FB15K, please run the following commands.
#################################### WN18RR ####################################
# RQE
python train_models.py --model RQE --dataset WN18RR --train_times 40000 --nbatches 10 --alpha 0.02 --dimension 300 --lmbda 0.25 --lmbda_two 0.25 --ent_neg_rate 2 --valid_step 2000
# HRQE
python train_models.py --model HRQE --dataset WN18RR --train_times 50000 --nbatches 10 --alpha 0.1 --dimension 300 --lmbda 0.3 --lmbda_two 0.01 --ent_neg_rate 1 --valid_step 2000
#################################### FB15K237 ####################################
# RQE
python train_models.py --model RQE --dataset FB15K237 --train_times 8000 --nbatches 100 --alpha 0.02 --dimension 500 --lmbda 0.5 --lmbda_two 0.01 --ent_neg_rate 10 --valid_step 400
# HRQE
python train_models.py --model HRQE --dataset FB15K237 --train_times 5000 --nbatches 100 --alpha 0.05 --dimension 500 --lmbda 0.5 --lmbda_two 0.01 --ent_neg_rate 10 --valid_step 400
#################################### WN18 ####################################
# RQE
python train_models.py --model RQE --dataset WN18 --train_times 4000 --nbatches 10 --alpha 0.04 --dimension 300 --lmbda 0.03 --lmbda_two 0.0 --ent_neg_rate 10 --valid_step 400
# HRQE
python train_models.py --model HRQE --dataset WN18 --train_times 8000 --nbatches 10 --alpha 0.05 --dimension 300 --lmbda 0.05 --lmbda_two 0.01 --ent_neg_rate 10 --valid_step 1000
#################################### FB15K ####################################
# RQE
python train_models.py --model RQE --dataset FB15K --train_times 2000 --nbatches 100 --alpha 0.02 --dimension 400 --lmbda 0.05 --lmbda_two 0.0 --ent_neg_rate 10 --valid_step 200
# HRQE
python train_models.py --model HRQE --dataset FB15K --train_times 4000 --nbatches 100 --alpha 0.02 --dimension 400 --lmbda 0.05 --lmbda_two 0.0 --ent_neg_rate 10 --valid_step 400This code is based on the OpenKE project.
We refer to the code of QuatE. Thanks for their contributions.
This is the code for paper "Learning Hierarchy-Aware Quaternion Knowledge Graph Embeddings with Representing Relations as 3D Rotations". If it helps your work, please cite the following paper:
@inproceedings{yang2022learning,
title={Learning Hierarchy-Aware Quaternion Knowledge Graph Embeddings with Representing Relations as 3D Rotations},
author={Yang, Jinfa and Ying, Xianghua and Shi, Yongjie and Tong, Xin and Wang, Ruibin and Chen, Taiyan and Xing, Bowei},
booktitle={Proceedings of the 29th International Conference on Computational Linguistics},
pages={2011--2023},
year={2022}
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent