Awesome

Context-aware Entity Typing in Knowledge Graphs

This is the source code for: Context-aware Entity Typing in Knowledge Graphs.

Requirements

• Python 3
• PyTorch >= 1.6.0
• dgl >= 0.5.3

Usage

Data preprocessing:

cd data
python preprocess.py --dataset FB15kET
python preprocess.py --dataset YAGO43kET

Train:

########### FB15kET ###########
# CET
python run.py --model CET --dataset FB15kET --load_ET --load_KG --neighbor_sampling \
--hidden_dim 100 --temperature 0.5 --lr 0.001 --loss FNA --beta 4.0 --cuda

# R-GCN
python run.py --model RGCN --dataset FB15kET --load_ET --load_KG --neighbor_sampling \
--hidden_dim 100 --lr 0.001 --loss FNA --beta 3.0 --cuda

# CompGCN
python run.py --model CompGCN --dataset FB15kET --load_ET --load_KG --neighbor_sampling \
--hidden_dim 100 --lr 0.001 --loss FNA --activation relu --cuda

########### YAGO43kET ###########
# CET
python run.py --model CET --dataset YAGO43kET --load_ET --load_KG --neighbor_sampling \
--hidden_dim 100 --temperature 0.5 --lr 0.001 --loss FNA --beta 2.0 --cuda

# R-GCN
python run.py --model RGCN --dataset YAGO43kET --load_ET --load_KG --neighbor_sampling \
--hidden_dim 100 --lr 0.001 --loss FNA --beta 2.0 --cuda

# CompGCN
python run.py --model CompGCN --dataset YAGO43kET --load_ET --load_KG --neighbor_sampling \
--hidden_dim 100 --lr 0.001 --loss FNA --activation relu --cuda

The KGE baselines can be found in KGE_baselines.

Acknowledgement

We refer to the code of <a href='https://github.com/dmlc/dgl'>DGL</a>. Thanks for their contributions.

Citation

If you find this code helpful, please kindly cite the following paper.

@inproceedings{pan-etal-2021-context-aware,
    title = "Context-aware Entity Typing in Knowledge Graphs",
    author = "Pan, Weiran and Wei, Wei and Mao, Xian-Ling",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2021",
    month = nov,
    year = "2021",
    address = "Punta Cana, Dominican Republic",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.findings-emnlp.193",
    doi = "10.18653/v1/2021.findings-emnlp.193",
    pages = "2240--2250",
}