There is a newer version of the record available.

Published June 2, 2026 | Version v1.1-submission

Korea Victimization-Risk Persona Benchmark (KVRB): survey-anchored synthetic personas with crime-prevention behavioral attributes

Authors/Creators

  • 1. Korean National Police Agency, AI Policy Division; KAIST

Description

A survey-anchored synthetic-persona dataset and reproducibility package that augments NVIDIA Nemotron-Personas-Korea with six crime-prevention behavioral attributes (financial vulnerability, digital literacy, authority deference, social isolation, prior victimization, reporting propensity). Each attribute is sampled from a weighted Korean national-survey conditional, so attribute distributions match the source surveys by construction. The Route-B build adds shared-demographic conditioning (age x sex x education) and one observed within-survey joint (victimization x reporting, from KCVS), with all inter-attribute structure labeled as measured or assumed.

Contents: construction and validation code (src/), aggregate survey anchor tables with provenance (config/anchors/), technical-validation outputs and figures (results/), methodology docs, and the data-descriptor manuscript draft. Not included: the 1,000,000-row enriched persona file (reproducible from the code + anchors) and the raw survey microdata (held under the source agencies' terms).

Synthetic only: no real victims, suspects, or PII. See ETHICS.md for the dual-use statement.

Notes

Derivative of NVIDIA Nemotron-Personas-Korea (CC-BY-4.0). Behavioral anchors are aggregate distributions derived from Korean national surveys (KGSS, KCVS, 가계금융복지조사, 디지털정보격차 실태조사, 노인실태조사 2023); only aggregate tables are redistributed, no microdata.

Files

kncop0526-star/korea-victimization-risk-benchmark-v1.1-submission.zip

Files (2.2 MB)

Additional details