Multiagent Data Mining for Enhanced Machine Learning Prediction of Delayed Fluorescence Materials

Area of Science:

• Materials Science
• Artificial Intelligence
• Organic Electronics

Background:

• Data acquisition for AI models in organic light-emitting diodes (OLEDs) is a major bottleneck.
• Limited experimental data hinders understanding of materials mechanisms and AI-driven discovery.

Purpose of the Study:

• To develop a multiagent AI framework for autonomous data extraction from PDF literature.
• To curate a comprehensive dataset of thermally activated delayed fluorescence (TADF) materials.
• To improve the accuracy of AI models for OLED material screening.

Main Methods:

• A multiagent AI framework was designed to extract molecular structures, properties, and experimental conditions from PDFs.
• The framework integrates specialized AI agents for efficient and accurate data retrieval.
• Data from over 1600 publications on TADF materials was curated and released publicly.

Main Results:

• Trained machine learning models achieved high accuracy in predicting six key properties for OLED material screening.
• The framework successfully extracted detailed measurement conditions and device architectures.
• The study demonstrated that improved data acquisition significantly enhances AI model predictive performance.

Conclusions:

• The developed AI framework lowers the barrier to accessing OLED-related data.
• This accelerates AI-driven material discovery by enhancing the accuracy of emitter screening models.
• Publicly releasing the curated dataset facilitates further research in OLED materials.