kartoun/Bio_ClinicalBERT_for_Alcohol_Use_Classification

Acknowledgment: This project utilizes the dataset and fine-tuned model developed by Dr. Uri Kartoun (https://urikartoun.com/).

Overview: This repository hosts the fine-tuned model, adapted specifically for the detection of alcohol use expressions in clinical narratives. This fine-tuned model is based on 1,000 simulated expressions, labeled as either 'inappropriate use of alcohol' or 'no use or acceptable use of alcohol'. It may serve particularly for studies that need to consider alcohol consumption as a significant covariate, such as those excluding patients from cohorts in liver disease research.

Model Description: The base model, emilyalsentzer/Bio_ClinicalBERT, has been fine-tuned to better recognize and categorize expressions related to alcohol use. This adaptation makes it highly suited for parsing and understanding nuanced medical texts where alcohol use status is relevant.

Performance: The fine-tuned model demonstrates high accuracy in classifying alcohol-related expressions, ensuring that its application in research and clinical settings is both reliable and effective.

Classification performance using a held-out set:

Getting Started: To use or further fine-tune the model with your own dataset of clinical expressions, please refer to the source code: https://github.com/kartoun/alcohol_use_classification_llms. The code provides all necessary instructions to replicate the fine-tuning process or to adapt it to new datasets potentially drawn from real healthcare systems.