ClaimGraph-BT
Claim extraction and cross-paper comparison for brain tumor AI literature.
Research focus
This project investigates how to extract claims from brain tumor AI papers and compare them across publications. I trained and evaluated context-enhanced biomedical claim detectors using paper-level splits and manually verified test data. PubMedBERT Context achieved 0.8391 macro F1 and 0.9301 AUC.
For structured claim extraction, deterministic repair and grounding increased macro field-presence F1 from 0.1668 to 0.4850.
The paper was accepted at AIST 2026, CCIS track. See the publication details. Related data: BrainTumorClaims.