This is the official repository for the paper "Classification of health product defect reports by deep learning".
- Version of software:
-
Anaconda version: 2.3.1
-
Python version: 3.7.15
-
CUDA version: 11.0.167
- The packages, dependencis, and version information required to run the provided notebook are included in requirements.txt file.
The Product_defects_notebook.ipynb file contains the full pipeline of this study.
The pipeline includes the following components:
- Data preprocessing
- Model training (loading, fine-tuning, and prompt-tuning)
- Model evaluation
- Interpretability analyses
- Performance analyses
The two model weights can be accessed and downloaded from this link:
- Bert-base fine-tuned model: MedDefects-BERT
- Bert-base deep-prompt-tuned model: MedDefects-DPT-BERT
Demo scripts and test datasets have been provided (15-Sep-2023) in the Demo folder. Please refert to the Readme.md file of that folder for more details and follow the instruction in the guide to run the scripts and reproduce the results.
If there are any questions, please contact: vicente.enrique@synapxe.sg