psgrghvuo/pubmedbert_bc5cdr

Model

This model is a fine-tuned version of microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext on BC5CDR dataset. It achieves the following results on the evaluation set:

• Loss: 0.0876
• Precision: 0.8882
• Recall: 0.9258
• F1: 0.9066

Examples from BC5CDR (Test Set)

All the entities in following examples are correctly predicted by the model:

• The authors report on six cases of famotidine - associated delirium in hospitalized patients who cleared completely upon removal of famotidine . The pharmacokinetics of famotidine are reviewed , with no change in its metabolism in the elderly population seen . The implications of using famotidine in elderly persons are discussed.
• Scleroderma renal crisis ( SRC ) is a rare complication of systemic sclerosis ( SSc ) but can be severe enough to require temporary or permanent renal replacement therapy . Moderate to high dose corticosteroid use is recognized as a major risk factor for SRC.

Model fails to extract all the entities from the following examples (correct Chemical entities are highlighted with lime colour and Disease entities with yellow colour):

• Famotidine is a histamine H2 - receptor antagonist used in inpatient settings for prevention of stress ulcers and is showing increasing popularity because of its low cost.
• We used high - resolution MRI and surface - based computational image analyses to map regional abnormalities in the cortex , hippocampus , white matter , and ventricles in 22 human subjects who used MA and 21 age - matched , healthy controls . Cortical maps revealed severe gray - matter deficits in the cingulate , limbic , and paralimbic cortices of MA abusers had 7.8% smaller hippocampal volumes than control subjects and significant white - matter hypertrophy.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 42
• num_epochs: 6

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.1+cu117
• Tokenizers 0.13.2