BioASQ-yesno-PudMedBERT

BioASQ Yes/No Question Classifier

---

Model Details

• Model architecture: BERT
• Pretrained base: microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext
• Fine-tuned on: BioASQ Phase B Yes/No question dataset
• Task type: Binary classification (Yes/No)
• Input format: Concatenated question and supporting context passages
• Output: Probability distribution over two classes ("Yes", "No")
• Tokenizer: Depends on base model (WordPiece or SentencePiece)

---

Dataset

• Name: BioASQ Task B Phase B Yes/No dataset
• Domain: Biomedical question answering
• Data format: Each sample consists of a yes/no question paired with one or more relevant context snippets extracted from biomedical abstracts
• Split: Standard train/dev split from BioASQ

---

Performance

Evaluation performed on the BioASQ dev set.

Usage Example

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("tmt3103/BioASQ-yesno-PudMedBERT")
model = AutoModelForSequenceClassification.from_pretrained("tmt3103/BioASQ-yesno-PudMedBERT")

def predict_yesno(question: str, context: str) -> str:
    inputs = tokenizer(question, context, truncation=True, padding=True, return_tensors="pt")
    outputs = model(**inputs)
    probs = torch.nn.functional.softmax(outputs.logits, dim=-1)
    return "Yes" if probs[0][1] > probs[0][0] else "No"

# Example usage
question = "Does aspirin reduce inflammation?"
context = "Aspirin is widely used as an anti-inflammatory medication in clinical practice."
print(f"Question: {question}\nPredicted answer: {predict_yesno(question, context)}")

---

Future Work & Maintenance

• Retrain regularly with updated BioASQ datasets to maintain relevance.
• Implement uncertainty estimation for safer decision support.
• Expand to multi-class or multi-label biomedical QA tasks.
• Optimize for deployment efficiency and latency reduction.

---

Contact & Support

For questions, issues, or collaboration inquiries, please contact:

• Author / Maintainer: Minh Tien
• Email: [email protected]
• GitHub: TMTien31

---