GrapheneChat: A Large Language Model for Enhancing Graphene Research.

Area of Science:

• Materials Science
• Nanotechnology
• Artificial Intelligence

Background:

• Graphene research spans multiple disciplines, presenting challenges in knowledge integration and literature retrieval.
• Existing methods require specialized expertise and efficient tools for navigating vast scientific information.

Purpose of the Study:

• To develop the first fine-tuned large language model (LLM) specifically for graphene research.
• To enhance domain-specific reasoning, experimental design, and literature knowledge retrieval in the field of graphene.

Main Methods:

• GrapheneChat was developed using a two-stage fine-tuning strategy: supervised fine-tuning (SFT) and direct preference optimization (DPO).
• A retrieval-augmented generation (RAG) framework was integrated for literature-grounded responses and reference support.
• The model was trained on comprehensive datasets of graphene-related monographs and scholarly articles.

Main Results:

• GrapheneChat demonstrated enhanced domain-specific reasoning and experimental design capabilities.
• The model provides literature-grounded responses with reference support, facilitating knowledge retrieval.
• Quantitative evaluations using GrapheneBench showed 91% accuracy, comparable to GPT-4, with lower computational costs.

Conclusions:

• GrapheneChat serves as an intelligent research assistant, facilitating interdisciplinary innovation in graphene science.
• The development establishes a paradigm for domain-specific LLMs to boost expert productivity in scientific literature mining.