glp1-ade-sentence-biomedbert

Sentence classifier: does the sentence describe an adverse drug event? Fine-tuned BiomedBERT on ADE Corpus v2 (deduplicated split, notebook 12). Test F1 0.90 [0.89–0.92], PR-AUC 0.96; at the ≥90%-recall threshold (0.58) precision 0.89 (TF-IDF baseline: 0.47). Limits: case-report English from older literature (domain shift); use as a triage aid for human pharmacovigilance review, never as the decision.

Part of GLP-1 Launch Intelligence by SwagataBhowmik. Public and synthetic data only; not affiliated with any pharmaceutical company; not medical advice.