Awesome

DisenKGAT

<h1 align="center"> DisenKGAT </h1> <h4 align="center">DisenKGAT: Knowledge Graph Embedding with Disentangled Graph Attention Network</h4> <h2 align="center"> Overview of DisenKGAT <img align="center" src="./fig/DisenKGAT_model.jpg" alt="..."> </h2>

This is our PyTorch implementation for the paper DisenKGAT-2021CIKM:

Junkang Wu, Wentao Shi, Xuezhi Cao, Jiawei Chen, Fuzheng Zhang, Wei Wu, Xiangnan He (2021) DisenKGAT: Knowledge Graph Embedding with Disentangled Graph Attention Network paper link. In CIKM2021.

Dependencies

Compatible with PyTorch 1.8 and Python 3.6.
Dependencies can be installed using requirements.txt.

Dataset:

We use FB15k-237 and WN18RR dataset for knowledge graph link prediction.
FB15k-237 and WN18RR are included in the data directory.

Training model:

Install all the requirements from requirements.txt.
Execute sh ./preprocess.sh for extracting the dataset and setting up the folder hierarchy for experiments.

Commands for reproducing the reported results on link prediction:

##### with TransE Score Function
# DisenKGAT (Composition: Subtraction)
python run.py -epoch 1500 -name TransE_sub_K2_D200 -model disenkgat\
      -hid_drop 0.1 -gcn_drop 0.3 \
      -score_func transe -opn sub -data FB15k-237  \
      -init_dim 200 -mi_train -mi_method club_s -mi_drop

# DisenKGAT (Composition: Multiplication)
python run.py -epoch 1500 -name TransE_sub_K2_D200 -model disenkgat\
      -hid_drop 0.1 -gcn_drop 0.3 \
      -score_func transe -opn mult -data FB15k-237  \
      -init_dim 200 -mi_train -mi_method club_s -mi_drop

# DisenKGAT (Composition: Crossover Interaction)
python run.py -epoch 1500 -name TransE_sub_K2_D200 -model disenkgat\
      -hid_drop 0.1 -gcn_drop 0.3 \
      -score_func transe -opn cross -data FB15k-237  \
      -init_dim 200 -mi_train -mi_method club_s -mi_drop

##### with DistMult Score Function
# DisenKGAT (Composition: Subtraction)
python run.py -epoch 1500 -name Distmult_sub_K2_D200 -model disenkgat\
      -score_func distmult -opn sub -gcn_layer 2 -batch 128 -test_batch 128 
      -early_stop 66 -num_factors 2 -mi_train -mi_method club_s

# DisenKGAT (Composition: Multiplication)
python run.py -epoch 1500 -name Distmult_sub_K2_D200 -model disenkgat\
      -score_func distmult -opn mult -gcn_layer 2 -batch 128 -test_batch 128 
      -early_stop 66 -num_factors 2 -mi_train -mi_method club_s

# DisenKGAT (Composition: Crossover Interaction)
python run.py -epoch 1500 -name Distmult_sub_K2_D200 -model disenkgat\
      -score_func distmult -opn cross -gcn_layer 2 -batch 128 -test_batch 128 
      -early_stop 66 -num_factors 2 -mi_train -mi_method club_s

##### with InteractE Score Function
# DisenKGAT (Composition: Subtraction)
python run.py -epoch 1500 -name InteractE_sub_FB15k_K2_D200_club_b_mi_drop -model disenkgat\
      -score_func interacte -opn sub -num_factors 2 \
      -mi_train -mi_method club_b -mi_drop

# DisenKGAT (Composition: Multiplication)
python run.py -epoch 1500 -name Mult_InteractE_FB15k_K2_D200_club_b_mi_drop -model disenkgat\
      -score_func interacte -opn mult -num_factors 2 \
      -mi_train -mi_method club_b -mi_drop

# DisenKGAT (Composition: Crossover Interaction)
python run.py -epoch 1500 -name InteractE_FB15k_K3_D200_club_b_mi_drop -model disenkgat\
      -score_func interacte -opn cross -num_factors 3 \
      -mi_train -mi_method club_b -mi_drop

##### Overall BEST:
python run.py -epoch 1500 -name InteractE_FB15k_K3_D200_club_b_mi_drop -mi_train -mi_drop

Citation

If you want to use our codes in your research, please cite:

@inproceedings{wu2021disenkgat,
  title={DisenKGAT: Knowledge Graph Embedding with Disentangled Graph Attention Network},
  author={Wu, Junkang and Shi, Wentao and Cao, Xuezhi and Chen, Jiawei and Lei, Wenqiang and Zhang, Fuzheng and Wu, Wei and He, Xiangnan},
  booktitle={Proceedings of the 30th ACM International Conference on Information \& Knowledge Management},
  pages={2140--2149},
  year={2021}
}

Acknowledgement

The project is built upon COMPGCN

For any clarification, comments, or suggestions please create an issue or contact me.