lrei/rad-small

This is a distilroberta-base model fined tuned to classify text into 3 categories:

• Rare Diseases
• Non-Rare Diseases
• Other

The details of how this model was built and evaluated are provided in the article:

Rei L, Pita Costa J, Zdolšek Draksler T. Automatic Classification and Visualization of Text Data on Rare Diseases. Journal of Personalized Medicine. 2024; 14(5):545. https://doi.org/10.3390/jpm14050545

@Article{jpm14050545,
AUTHOR = {Rei, Luis and Pita Costa, Joao and Zdolšek Draksler, Tanja},
TITLE = {Automatic Classification and Visualization of Text Data on Rare Diseases},
JOURNAL = {Journal of Personalized Medicine},
VOLUME = {14},
YEAR = {2024},
NUMBER = {5},
ARTICLE-NUMBER = {545},
URL = {https://www.mdpi.com/2075-4426/14/5/545},
PubMedID = {38793127},
ISSN = {2075-4426},
DOI = {10.3390/jpm14050545}
}

Note that the in the article the larger roberta-base model is fine-tuned instead. This is a smaller model. This model is shared for demonstration and validation purposes. Hyper-parameters were not tuned.

Using this model

Simplest way to use this model is via a huggingface transformers' pipeline.

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-classification", model="lrei/rad-small")

# Simple high-level usage
pipe(["The patient suffer from a complex genetic disorder.", "The patient suffers from a common genetic disorder."])

Dataset

The dataset used to train this model is available on zenodo. It is a subset of abstracts obtained from PubMed and sorted into the 3 classes on the basis of their MeSH terms.

Like the model, the dataset is provided for demonstration and methodology validation purposes. The original PubMed data was randomly under-sampled.

Code

The code used to create this model is available on Github.

Test Results

Averaged over all 3 classes:

average	precision	recall	F1
micro	0.84	0.84	0.84
macro	0.84	0.84	0.84