Type-enhanced Inductive Knowledge Completion

The repository provides the code and data used in our experiments.

Requirements

python==3.7.13

torch==1.5.0

dgl==0.4.2

scikit-learn

tqdm

lmdb

Directory

data: The inductive datasets split by GraIL

types: The raw types of entities we obtained and the types of entities after preprocessing.

expri_save_models: The trained models to generate experimental results in the paper.

Run

We provide the commands to train and test our model, and the illustration of their parameters. Take nell_v1 for example.

• training python train.py -d nell_v1 -e nell_v1 -ne 20 --ont

  • -d: the name of training dataset
  • -e: the directory of saved models
  • -ne: the number of epoches
  • --ont: type-enhanced model

• test on AUC-PR python test_auc.py -d nell_v1_ind -e nell_v1 --ont --runs 5

  • -d: the name of test dataset
  • -e: the directory of saved models
  • --ont: type-enhanced model
  • --runs: run times

• test on Hits@10 python test_ranking.py -d nell_v1_ind -e nell_v1 --ont

  • -d: the name of test dataset
  • -e: the directory of saved models
  • --ont: type-enhanced model