Overview of the RelEns.
Code for the paper Relation-aware Ensemble Learning for Knowledge Graph Embedding. Accepted by EMNLP 2023. This paper proposes a relation-aware ensemble method with divide-concur method and achieving state-of-the-art KG embedding performance.
Here, we provide a quick start guide on how to reproduce the results.
For the OGB dataset, you don't need to do anything. We will download it automatically. For WN18RR, FB15k-237 and NELL-995, you can download them from here
git clone https://github.com/LARS-research/RelEns.git
cd RelEns
pip install -r requirements.txt
You need to prepare the predictions for the base model.
The directory structure is as follows:
|-- ranks
|-- {dataset1_name}
|-- {model1_name}_valid_ranks.npy
|-- {model1_name}_test_ranks.npy
|-- {dataset2_name}
|-- {model2_name}_valid_ranks.npy
|-- {model2_name}_test_ranks.npy
For WN18RR, FB15k-237, and NELL-995, the shape of the ranks npy file is (#number of samples, #number of entities) and contains the entity ranks for every sample. The rank belongs to the range [1, #number of entities].
For OGB datasets, the shape of the ranks npy file is (#number of samples, 1002). It includes 1002 columns, which consist of 1 positive head rank, 500 head negative ranks, 1 positive tail rank, and 500 tail negative ranks.
To run the code, you can execute the ogb_main.ipynb file for OGB datasets and main.ipynb for other datasets. Additionally, the log for ogbl-biokg and WN18RR are already available.
Please kindly cite this paper if you use the code.
@article{yue2023relation,
title={Relation-aware Ensemble Learning for Knowledge Graph Embedding},
author={Yue, Ling and Zhang, Yongqi and Yao, Quanming and Li, Yong and Wu, Xian and Zhang, Ziheng and Lin, Zhenxi and Zheng, Yefeng},
year={2023},
eprint={2303.08917},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
RelEns is released under the MIT license. Further details can be found here.