Cognitively Layered Data Synthesis for Domain Adaptation of LLMs to Space Situational Awareness

Researchers developed BD-FDG, a framework for adapting large language models to complex engineering domains like space situational awareness. The method creates high-quality training datasets using structured knowledge organization and cognitive layering, resulting in SSA-LLM-8B that shows 144-176% BLEU-1 improvements while maintaining general performance.

• →BD-FDG framework addresses key challenges in adapting LLMs to specialized engineering domains through structured knowledge trees and cognitive layering.
• →The approach generates training data across nine categories and six cognitive levels from basic recall to creative problem-solving.
• →SSA-LLM-8B achieved 144-176% relative BLEU-1 improvements on domain-specific tasks while preserving general benchmark performance.
• →The framework constructed a 230K sample dataset for space situational awareness applications.
• →Results demonstrate an 82.21% win rate over baseline models in arena comparisons for domain-specific tasks.