nawazishpatana/claim-detector-brain-tumor

Claim Detector for Brain Tumor Research Papers

Model: SciBERT+Context | F1: 0.8003

Overview

Detects claim sentences in brain tumor research paper abstracts.

Performance

• —F1: 0.8003
• —Accuracy: 0.8974
• —Recall: 0.8117

Dataset

• —15,785 sentences from 1,496 papers
• —Claims: 3,997 (25.3%), Non-claims: 11,788 (74.7%)
• —Split: 70/15/15 train/val/test

Quick Start

from transformers import AutoTokenizer, AutoModel
import torch

model_name = "nawazishpatana/claim-detector-brain-tumor"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModel.from_pretrained(model_name)

text = "Our model achieved 95% accuracy."
inputs = tokenizer(text, return_tensors="pt")
with torch.no_grad():
    outputs = model(**inputs)

Ablations

1. 1.SciBERT+Context: F1=0.8003
2. 2.SciBERT: F1=0.7483
3. 3.PubMedBERT+Context: F1=0.7813

Training Config

• —Optimizer: AdamW (LR backbone=2e-05, head=0.001)
• —Loss: CrossEntropyLoss + label smoothing=0.1
• —Early stopping: patience=3
• —Context: Yes