Source code for our paper: Incorporating Global Information in Local Attention for Knowledge Representation Learning

• conda
• pytorch (version >= 1.1.0)

The four public benchmark datasets for link prediction experiments with their folder names are given below.

• Freebase: FB15k-237
• Nell: NELL-995
• Kinship: kinship
• UMLS: umls

When running for first time:

 $ sh prepare.sh

Then reproducing the results in the paper by:

• Nell

   $ python3 main.py --get_2hop True --use_2hop True
  

• Kinship

   $ python3 main.py --data ./data/kinship/ --output_folder ./checkpoints/kinship/out/ --lr 1e-2 --epochs_gat 4000 --epochs_conv 400 --batch_size_gat 8544 --drop_GAT 0.3 --weight_decay_conv 1e-5 --valid_invalid_ratio_conv 10 --out_channels 50 --drop_conv 0.0 --get_2hop True --use_2hop True
  

• Other datasets

Parameters of other datasets are given in appendix of the paper.

For any comments or suggestions, please contact zhhan@connect.hku.hk