DDIE-KGE-MFL

Source code for "Extracting Drug-drug Interactions from Biomedical Texts using Knowledge Graph Embeddings and Multi-focal Loss" CIKM'22.

Implementation setup

Our project implementation depends on the following environments.

• Python == 3.7
• Pytorch == 1.9.0
• Transformers == 4.3.2
• Stanza == 1.3.0

you need to run download_stanza.sh to get the stanza models firstly and then to train or evaluate.

Train

1. Download PubMedBERT(config.json,pytorch_model.bin,vocab.txt) and put them to pubmedbert directory.
2. Then execute run.sh to train the model

Evaluation

1. Download our pretrained model from GoogleDrive to pretrained directory and decompress model.tar.gz or put your pretrained model into pretrained directory.
2. Execute eval.sh to test the model on test dataset.