TOPPAN, NTT Data, and FPSC Advance Large-Scale Dataset Preparation for Digital Agency's 'Government AI Gen'nai'

TOPPAN Inc. (Headquarters: Bunkyo-ku, Tokyo; President and CEO: Haruhiko Noguchi; hereinafter “TOPPAN”), a group company of TOPPAN Holdings, NTT Data Corporation (Headquarters: Koto-ku, Tokyo; President and CEO: Masanori Suzuki; hereinafter “NTT Data”), and Fides Policy & Strategy Consulting Inc. (Headquarters: Shibuya-ku, Tokyo; Representative Director: Yusuke Shimizu; hereinafter “FPSC”) will leverage their respective expertise to prepare large-scale datasets for use in the 'Government AI Gen'nai' (※), implement AI-Ready processing (structuring data with quality assurance for AI comprehension), and examine AI-targeted dataset standards as part of the Digital Agency’s “Survey, Collection, and Processing Project for Large-Scale Datasets for Government AI” (hereinafter “this project”).

This project aims to collect and process cross-governmental useful data (hereinafter “common government data”) and data used in specific operations by individual ministries and agencies (hereinafter “domain-specific data”), preparing them in formats suitable for AI use, while also examining standards for AI-targeted datasets, in support of the government’s initiative to utilize 'Government AI Gen'nai'.

TOPPAN has been commissioned by the Digital Agency for this project and, in collaboration with NTT Data and FPSC, will implement AI-Ready processing on approximately 18 million pages (estimated) of common government and domain-specific data, working to establish a data infrastructure that supports the utilization of 'Government AI Gen'nai'.

Figure: Concept of this project

Background of the Project

In Japan, administrative service efficiency is increasingly required due to population decline and aging. Amid this, expectations for AI are rising. The Digital Agency is promoting 'Government AI Gen'nai' as a starting point for societal implementation of government AI use. The 'Government AI Gen'nai' environment is being developed to allow government employees to safely and appropriately use AI, with large-scale demonstrations planned for approximately 180,000 government employees by fiscal year 2026.

To expand the use of 'Government AI Gen'nai', it is essential to prepare high-quality datasets that AI can reference and learn from. Recently, in improving AI response accuracy, the quality of training data and data referenced in Retrieval-Augmented Generation (RAG) has become as important as improvements in AI model performance. Therefore, in addition to data collection, metadata preparation and data structure standardization are required.

Furthermore, from the perspectives of economic security and data sovereignty, there is a growing need to establish a foundation that enables the conversion of sensitive specialized knowledge data, such as administrative documents, into high-quality AI-Ready data within a secure domestic environment, and to manage and utilize it appropriately. In response, the Digital Agency is advancing the preparation of large-scale datasets for 'Government AI Gen'nai', and TOPPAN, NTT Data, and FPSC will work through this project to AI-Ready common government data and examine AI-targeted dataset standards.

Overview of the Project

This project will collect and process common government data and domain-specific data, prepare metadata, and examine AI-targeted dataset standards, based on specifications provided by the Digital Agency.

1) Rights Processing

Negotiate usage rights and permissions for intellectual property and other rights related to collected data.

2) Collection and Processing of Common Government and Domain-Specific Data

Collect common government data (e.g., cabinet decisions, Diet records, white papers) and domain-specific data (e.g., books used in specific operations by ministries and agencies). Additionally, perform preprocessing such as OCR and data cleaning to convert the data into high-quality formats suitable for AI search, learning, and inference.

3) Design, Creation, and Assignment of Metadata for Collected Datasets

Design, create, and assign metadata—including provenance, quality, rights, and usage conditions—for datasets collected through this project. This will establish an environment where AI systems and users can appropriately assess data reliability and usability.

4) Examination of the 'Government AI Gen'nai Dataset Standard'

Through expert interviews and discussion panels, examine structuring of AI-targeted datasets and metadata specifications, and draft a preliminary version of a government-wide AI-targeted dataset standard applicable across administrative agencies.

【Roles of Each Company】

・TOPPAN

Leveraging business analysis capabilities and data processing and structuring technologies developed through BPO operations in the public and financial sectors, TOPPAN will handle data selection and collection based on administrative use cases, rights processing, OCR processing, and data preprocessing.

Additionally, TOPPAN will AI-Ready administrative documents and prepare datasets suitable for search and inference.

・NTT Data

Drawing on expertise gained from digital archives for government agencies and support for generative AI implementation, NTT Data will handle metadata design, creation, and assignment, and examine government-wide AI-targeted dataset standards. By preparing metadata on data provenance and usage conditions, NTT Data will support the construction of a reliable data infrastructure.

・FPSC

Leveraging experience as administrative officials in central government agencies, including legal affairs and inter-ministerial consultations, FPSC will support the organization of legal issues related to AI data utilization and the development of rights processing frameworks. FPSC will assist in intellectual property rights discussions, including copyright, and rights processing operations, contributing to appropriate and smooth data utilization.

Future Outlook

The three companies will complete the establishment of a common government data infrastructure for 'Government AI Gen'nai', used by government employees, by March 2027.

Additionally, through expert interviews and discussion panels, they will continue to examine dataset structuring and metadata specifications for AI, contributing to the development of data infrastructure supporting the utilization of 'Government AI Gen'nai' promoted by the Digital Agency.

Going forward, the three companies will continue to leverage their expertise in data collection and processing, metadata preparation optimized for AI, and data utilization to support the advancement of AI use by governments, local authorities, and enterprises, contributing to the realization of a digital society.

※ 'Government AI Gen'nai': A generative AI usage environment developed in-house by the Digital Agency

* Product, service, company, and organization names mentioned in this press release are trademarks or registered trademarks of their respective owners.

* The content of this press release is current as of the date of announcement and is subject to change without prior notice.

End