BioBERT-DAGsHub/relation-extraction

BioBERT for Relation Extraction

To train an RE model with BioBERT-v1.1 (base), please use following command line: Before training, please run ./preprocess.sh to preprocess the datasets downloaded in biobert-pytorch (see here).

Additional Requirements

• sklearn: Used for RE evaluation (pip install scikit-learn)
• pandas : Used for RE evaluation (pip install pandas)

Training

export SAVE_DIR=./output
export DATA="GAD"
export SPLIT="1"
export DATA_DIR=../datasets/RE/${DATA}/${SPLIT}
export ENTITY=${DATA}-${SPLIT}

export MAX_LENGTH=128
export BATCH_SIZE=32
export NUM_EPOCHS=3
export SAVE_STEPS=1000
export SEED=1

python run_re.py \
    --task_name SST-2 \
    --config_name bert-base-cased \
    --data_dir ${DATA_DIR} \
    --model_name_or_path dmis-lab/biobert-base-cased-v1.1 \
    --max_seq_length ${MAX_LENGTH} \
    --num_train_epochs ${NUM_EPOCHS} \
    --per_device_train_batch_size ${BATCH_SIZE} \
    --save_steps ${SAVE_STEPS} \
    --seed ${SEED} \
    --do_train \
    --do_predict \
    --learning_rate 5e-5 \
    --output_dir ${SAVE_DIR}/${ENTITY} \
    --overwrite_output_dir

Evaluation

python ./scripts/re_eval.py --output_path=${SAVE_DIR}/${ENTITY}/test_results.txt --answer_path=${DATA_DIR}/test_original.tsv

To evaluate the prediction, please use scripts/re_eval.py file. For an example running script for 10-cv experiment, please task a look at run_re_10cv.sh and scripts/re_eval_10cv.sh.

Evaluation Results

BioBERT

	Precision (%)	Recall (%)	F1 (%)
GAD	77.09	88.50	82.37
EUADR	77.48	96.15	85.13

Contact

For help or issues using BioBERT-PyTorch, please create an issue and tag @wonjininfo.