BioBert-PubMed200kRCT
This model is a fine-tuned version of dmis-lab/biobert-base-cased-v1.1 on the PubMed200kRCT dataset. It achieves the following results on the evaluation set:
- Loss: 0.2832
- Accuracy: 0.8934
Model description
More information needed
Intended uses & limitations
The model can be used for text classification tasks of Randomized Controlled Trials that does not have any structure. The text can be classified as one of the following:
- BACKGROUND
- CONCLUSIONS
- METHODS
- OBJECTIVE
- RESULTS
The model can be directly used like this:
from transformers import TextClassificationPipeline
from transformers import AutoTokenizer, AutoModelForSequenceClassification
model = AutoModelForSequenceClassification.from_pretrained("pritamdeka/BioBert-PubMed200kRCT")
tokenizer = AutoTokenizer.from_pretrained("pritamdeka/BioBert-PubMed200kRCT")
pipe = TextClassificationPipeline(model=model, tokenizer=tokenizer, return_all_scores=True)
pipe("Treatment of 12 healthy female subjects with CDCA for 2 days resulted in increased BAT activity.")
Results will be shown as follows:
[[{'label': 'BACKGROUND', 'score': 0.0027583304326981306},
{'label': 'CONCLUSIONS', 'score': 0.044541116803884506},
{'label': 'METHODS', 'score': 0.19493348896503448},
{'label': 'OBJECTIVE', 'score': 0.003996663726866245},
{'label': 'RESULTS', 'score': 0.7537703514099121}]]
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 64
- eval_batch_size: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 2.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 0.3587 | 0.14 | 5000 | 0.3137 | 0.8834 |
| 0.3318 | 0.29 | 10000 | 0.3100 | 0.8831 |
| 0.3286 | 0.43 | 15000 | 0.3033 | 0.8864 |
| 0.3236 | 0.58 | 20000 | 0.3037 | 0.8862 |
| 0.3182 | 0.72 | 25000 | 0.2939 | 0.8876 |
| 0.3129 | 0.87 | 30000 | 0.2910 | 0.8885 |
| 0.3078 | 1.01 | 35000 | 0.2914 | 0.8887 |
| 0.2791 | 1.16 | 40000 | 0.2975 | 0.8874 |
| 0.2723 | 1.3 | 45000 | 0.2913 | 0.8906 |
| 0.2724 | 1.45 | 50000 | 0.2879 | 0.8904 |
| 0.27 | 1.59 | 55000 | 0.2874 | 0.8911 |
| 0.2681 | 1.74 | 60000 | 0.2848 | 0.8928 |
| 0.2672 | 1.88 | 65000 | 0.2832 | 0.8934 |
Framework versions
- Transformers 4.18.0.dev0
- Pytorch 1.10.0+cu111
- Datasets 1.18.4
- Tokenizers 0.11.6
Citing & Authors
If you use the model kindly cite the following work
@inproceedings{deka2022evidence,
title={Evidence Extraction to Validate Medical Claims in Fake News Detection},
author={Deka, Pritam and Jurek-Loughrey, Anna and others},
booktitle={International Conference on Health Information Science},
pages={3--15},
year={2022},
organization={Springer}
}