Protocol for Prospective Evaluation of Screening Algorithms integrated in NeutrinoReview
The study prospectively evaluates two LLM‑based screening algorithms, 5‑Tier and CAL‑X, integrated into the NeutrinoReview web application at CERN. Using 1,383 unique citations from a systematic review on indoor air quality in Italy, six screening runs were executed with the LLaMA 3.1‑8B model, yielding preliminary workload reductions between 36 % and 72 % while sensitivity assessment remains pending. Final analyses will compare algorithmic and human decisions, assess sensitivity and error patterns, and explore combined configurations, with all outputs time‑stamped and embargoed for transpare…
The study prospectively evaluates two LLM‑based screening algorithms, 5‑Tier and CAL‑X, integrated into the NeutrinoReview web application at CERN. Using 1,383 unique citations from a systematic review on indoor air quality in Italy, six screening runs were executed with the LLaMA 3.1‑8B model, yielding preliminary workload reductions between 36 % and 72 % while sensitivity assessment remains pending. Final analyses will compare algorithmic and human decisions, assess sensitivity and error patterns, and explore combined configurations, with all outputs time‑stamped and embargoed for transpare…
It demonstrates a Swiss‑hosted, transparent application of large‑language‑model technology in systematic review screening, informing Swiss AI policy on responsible deployment, trust, and real‑world validation.
Original source record. Open the original record to verify the underlying announcement.