PlaceBot evaluation datasets, retained model outputs and reproducibility materials

Le résumé fourni par la source

This record contains the benchmark datasets, retained model outputs, repeatability runs and reproducibility materials supporting the manuscript “PlaceBot: A Python tool for AI-assisted georeferencing of museum specimen locality data, with cross-vendor model evaluation”. The study evaluated 12 large language models using locality information from UK bumblebee (Bombus) and globally distributed dragonfly (Odonata) museum specimens. The available source pools contained 709 UK and 908 global records. Fixed cohorts comprising the first 700 UK records and first 900 global records in the retained input-file order were used for the main evaluation, producing 19,200 model-record attempts. Repeatability was evaluated using a fixed set of 200 records—100 UK and 100 global—with ten independent runs per dataset and 12 models, producing 24,000 model-record attempts. The deposit contains: (1) the benchmark input tables; (2) the 24 retained main model-output files; (3) the 20 complete repeatability-run JSON files; and (4) portable analysis scripts, derived source tables, model/run documentation, figure-generation code and the evaluated prompt. Rendered figures are presented in the associated paper and are not duplicated in this deposit. The source specimen records derive from the Natural History Museum, London, Collection specimens dataset: https://doi.org/10.5519/0002965.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.