Nemotron-Personas-Korea
lastmod 2026-04-27created 2026-04-272 edits
A dataset of 1 million synthetic personas based on the real population distribution of South Korea
About
- The first large-scale Korean-language persona dataset reflecting real demographic, geographic, and personality distributions, based on public data from Statistics Korea, the Supreme Court of Korea, the National Health Insurance Service, and others
- Contains 7 million personas in 1 million records, made up of 26 fields including name, sex, age, marital status, education level, occupation, and region of residence
- Built using NeMo Data Designer, an enterprise-grade synthetic data generation system, and the google/gemma-4-31B-it model
- Reflects older age groups, rural areas, and diverse education and occupation distributions more faithfully than existing persona datasets, helping mitigate bias in sovereign AI models
- Freely usable for both commercial and non-commercial purposes under the CC BY 4.0 license