The source code of Disconnected Emerging Knowledge Graph Oriented Inductive Link Prediction

Requirements

The required packages are listed in requirement.txt

The extended inductive link prediction experiments

All train and test data can be found in the data floder. Specifically, we train our model on train.txt in {dataset}_{version}. The main results are evaluated on the test.txt in {dataset}_{version}_mix, the results with enclosing links are evaluated on test.txtin {dataset}_{version}_enc, the results with bridging links are evaluated on test.txtin {dataset}_{version}_bri.

For example, to train the model DEKG-ILP on EQ of FB15k-237, run the following command:

python train.py -d FB15k-237_EQ -e DEKG-ILP_FB15k-237_EQ

To test DEKG-ILP, run the following commands:

# main result
python test_rank.py -d FB15k-237_EQ_mix -e DEKG-ILP_FB15k-237_EQ
# enclosing links only
python test_rank.py -d FB15k-237_EQ_enc -e DEKG-ILP_FB15k-237_EQ
# bridging links only
python test_rank.py -d FB15k-237_EQ_bri -e DEKG-ILP_FB15k-237_EQ

Acknowledgement

Our code refer to the code of Grail. Thanks for their contributions very much.