LLM-CD: A Cognitive Diagnosis Framework for Large Language Models(LLMs) ——Experimental Research Based on Green and Low-Carbon Knowledge

Large language models (LLMs) such as ChatGPT and DeepSeek have demonstrated significant potential for application across various fields. However, how to design appropriate application pathways for LLMs, especially whether they remain effective for domain-specific knowledge, is a question that urgently requires in-depth exploration. This study focuses on the domain of green and low-carbon knowledge, using LLMs as the research object and integrating nine cognitive diagnostic models from cognitive psychology to develop a cognitive diagnosis framework for LLMs (LLM-CD). This framework is capable of knowledge point extraction, multi-role generation, and automated assessment, and can quantify the LLMs' mastery of knowledge points. Based on this framework, we conducted an experimental study on the cognitive diagnosis of LLMs in the green and low-carbon domain. The results show that LLMs have significantly different levels of mastery over various green and low-carbon knowledge points, with higher proficiency in directly related concepts such as green, energy-saving, and low-carbon, compared to indirectly related concepts like water resources and air pollution. The experimental results can clearly identify the LLMs' cognitive status regarding different green and low-carbon knowledge points, thereby providing directional guidance for their application in this domain. Moreover, this framework can be transferred to other specific fields to facilitate the cross-empowerment of LLMs in multiple domains.