Inferring Gender from Author Names with Local LLMs: A Multi-Model Evaluation [Dataset]

This dataset accompanies the paper "Inferring Gender from Author Names with Local LLMs: A Multi-Model Evaluation" (Journal of Informetrics, 2026). It contains 194,528 first name–country pairs derived from Wikidata, along with gender predictions (Man / Woman / Unknown) produced by 25 open-weight large language models run locally via Ollama, ranging from 270M to 70B parameters. The ground truth is based on the gender distribution of real people registered in Wikidata for each name and country of origin. The dataset includes: (1) the ground-truth reference file with name, country, and sex; (2) the Wikidata source file with gender probability and frequency statistics; (3) one prediction file per model (25 total); and (4) Python scripts to reproduce the evaluation metrics (precision, recall, F1-score) and all figures reported in the paper. Data are released under CC BY 4.0; code under Apache 2.0