nawazishpatana/pubmedbert-main-claim-extraction

PubMedBERT for Main Claim Extraction

Model description

This model is a fine-tuned version of PubMedBERT for sentence-level scientific claim extraction in brain tumor / neuro-oncology abstracts.

The notebook then uses the predicted claim probability per sentence to rank sentences within each abstract and select the top-ranked sentence as the predicted main claim.

Dataset

The dataset was uploaded to:

nawazishpatana/brain-tumor-claim-extraction

Sentence-level evaluation results

• —Accuracy: 0.8641
• —Precision: 0.7166
• —Recall: 0.7922
• —F1-score: 0.7525

Base model

microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext