| ---
|
| language: en
|
| license: mit
|
| tags:
|
| - medical
|
| - classification
|
| - biobert
|
| - pubmedqa
|
| - healthcare-rag
|
| datasets:
|
| - qiaojin/PubMedQA
|
| metrics:
|
| - f1
|
| pipeline_tag: text-classification
|
| ---
|
|
|
| # BioBERT Medical Query Classifier
|
|
|
| Fine-tuned `dmis-lab/biobert-v1.1` for classifying medical questions into 6 categories.
|
|
|
| ## Categories
|
| | ID | Category |
|
| |----|----------|
|
| | 0 | Diagnosis |
|
| | 1 | General |
|
| | 2 | Medication |
|
| | 3 | Prevention |
|
| | 4 | Symptoms |
|
| | 5 | Treatment |
|
|
|
| ## Results
|
| | Metric | Score |
|
| |--------|-------|
|
| | Macro F1 | 0.9066 |
|
| | Weighted F1 | 0.9094 |
|
| | Accuracy | 0.9088 |
|
|
|
| ## Training Config
|
| | Item | Value |
|
| |------|-------|
|
| | Base model | dmis-lab/biobert-v1.1 |
|
| | Dataset | qiaojin/PubMedQA (211,186 rows) |
|
| | Split | 80/10/10 |
|
| | Epochs | 3 |
|
| | Learning rate | 2e-5 |
|
| | Batch size | 16 |
|
| | Class weights | Balanced (custom WeightedTrainer) |
|
|
|
| ## Usage
|
| from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
| import torch
|
|
|
| tokenizer = AutoTokenizer.from_pretrained("AbdoMatrix/biobert-medical-classifier")
|
| model = AutoModelForSequenceClassification.from_pretrained("AbdoMatrix/biobert-medical-classifier")
|
|
|
| text = "What are the symptoms of diabetes?"
|
| inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=256)
|
|
|
| with torch.no_grad():
|
| outputs = model(**inputs)
|
|
|
| predicted = model.config.id2label[torch.argmax(outputs.logits, dim=1).item()]
|
| print(predicted) # → Symptoms
|
|
|
| ## Project
|
| Healthcare RAG-Powered Medical Q&A Assistant
|
| eyouth x DEPI | Microsoft Machine Learning Track | 2026
|
| GitHub: https://github.com/AbdooMatrix/Healthcare-RAG-Powered-Medical-QA-Assistant
|
| |
