Nemotron-Personas-Finance-Japan: 日本の人口分布と金融コンテキストを捉える合成ペルソナデータセット

The Nemotron-Personas-Finance-Japan dataset, a financial domain extension of the open-source Nemotron-Personas-Japan persona dataset, has been released by Elith and NVIDIA. This 1 million record dataset combines Japanese official statistics, household financial conditions, and financial literacy to create a unique open persona dataset of its kind.

Built using the NeMo SDG-PGMs framework for probabilistic graphical model generation and the NeMo Data Designer for an enterprise synthetic data generation system, the dataset aims to support the development of region-specific AI without using real customer data. It supports various use cases from conversation agents to customer simulations and model evaluation, ensuring safe and practical financial AI development.

The dataset considers Japanese population statistics, household financial conditions, and financial literacy, generating completely synthetic personas with no real individuals involved to minimize privacy risks. It includes data from major Japanese financial institutions like insurance companies, banks, and fintech firms, enhancing its practicality for financial AI applications.

The dataset is available under the CC BY 4.0 license, making it accessible for both commercial and non-commercial use. It also complements the Nemotron-Personas-Japan dataset, offering 900,000 additional personas and expanding the range of industries to 50, promoting more granular data for Japanese language projects. The release also includes Nemotron-PII-Japan, a 100,000 record dataset containing labeled Japanese business documents for training and testing PII detection and masking models.

Developers can use these datasets to simulate customers and test the protection of personal information in their pipelines.