DMPBridge: Automating Data Management Plan Document Conversion Using Large Language Models

DMPBridge is an open-source project developed by the Fair Data Innovations Hub. Its core goal is to use large language model technology to automatically convert PDF-format Data Management Plans (DMPs) into structured JSON metadata compliant with the RDA common standards. It also supports DMPTool extension fields, solving the challenges of automated processing and cross-platform interoperability for traditional PDF-format DMPs, and providing an intelligent solution for research data management.

In modern scientific research, DMPs are an essential part of projects. However, the traditional PDF format is unstructured, making automated processing and cross-platform interoperability difficult. The RDA has introduced common DMP standards to establish unified metadata specifications, but a large number of historical DMPs are still in PDF format, and manual conversion is time-consuming and error-prone.

DMPBridge uses large language models to achieve automatic conversion from PDF to RDA-standard JSON. It supports DMPTool extension fields to ensure interoperability with mainstream tools. Using Jupyter Notebook as the development environment, the code has high readability and interactivity, making it easy to view the entire process of parsing, extraction, and conversion.

1. PDF Parsing: Use mature libraries to handle complex formats such as multi-column layouts and tables, accurately extracting text content;
2. LLM Content Understanding: Identify key information in DMPs such as data descriptions and storage strategies. Its semantic understanding capability is superior to traditional rule-based methods, adapting to differences between different templates;
3. Structured Output: Map to the RDA common standard JSON Schema, while supporting DMPTool extension fields to achieve seamless integration with existing infrastructure.

• Research institutions: Batch process historical DMPs to build a unified database;
• Funding agencies: Automate review processes and quickly extract and compare key information from DMPs;
• Interoperability research: Promote analysis of DMP quality, completeness, and compliance, and improve data management practices.

The project is released as open-source, lowering technical barriers and promoting community collaboration. The technology stack is practical: Jupyter Notebook reduces learning thresholds, and the Python ecosystem supports function expansion. The integration of large language models reflects the innovative application of AI in the field of research data management, allowing developers to conduct secondary development to adapt to specific needs.

DMPBridge solves the problem of DMP format interoperability and provides a path for the intelligent transformation of research data management. In the future, it is expected to support more document formats, complex semantic understanding tasks, and rich metadata standards, providing stronger technical support for the implementation of open science and FAIR data principles.