Task 3：Symptom Label Inference (SLI)

This dir contains the code of BERT-MTL model for SLI task.

The model utilizes the BIO labels and contains three training objectives:

• BIO tag prediction
• Symptom name classification
• Symptom label classification

The model is only valid for SLI-IMP task since it requires the annotated BIO tags. DialoAMC do not provide BIO tags for self-reports.

Requirements

• python>=3.5
• torch>=1.4.0
• transformers==2.7.0
• seqeval==1.2.2
• pytorch-crf==0.7.2
• tqdm==4.42.1
• pandas>=1.0.3

pip install -r requirements.txt

Preprocess

python preprocess.py

Training

python train.py --cuda_num 0

Inference

python inference.py --dataset test --cuda_num 0 

Evaluation

cd .. || exit
python eval_sli.py --target imp --gold ../../dataset/test.json --pred MTL-SLI/test_imp_pred.json