Improving LLM systematic reviews with BERT+GCN uncertainty signals
August 18, 2026
Integrating an auxiliary BERT+GCN classifier provides calibrated uncertainty signals that improve the efficiency of LLM-assisted systematic review screening. The study identifies optimal prompt-delivery strategies across eight drug-class datasets to maximize the benefit-to-cost ratio during title-abstract screening.
HOW THIS AFFECTS YOU
●
researcherYou can use structured uncertainty signals from hybrid models to calibrate LLM decision-making.
●
healthThis could make automated drug-class screening more reliable for medical literature reviews.