LMCD: Language Models are Zeroshot Cognitive Diagnosis Learners

TL;DR

LMCD tackles the cold-start cognitive diagnosis problem by leveraging large language models to diffusely enrich knowledge concepts and exercises (Knowledge Diffusion) and to fuse this enriched content with student cognitive states via a causal-attention based encoder (Semantic-Cognitive Fusion). The approach aligns the resulting representations with off-the-shelf cognitive diagnosis model heads, enabling prediction of responses $y_{uv}$ and estimation of student proficiency and item difficulty in a relative, student-specific manner. Extensive experiments on two real-world datasets show that LMCD consistently outperforms state-of-the-art graph- and NLP-based baselines in both exercise cold-start and cross-domain cold-start settings, with ablations validating the importance of diffusion content, causal fusion, and relative difficulty. The work demonstrates that integrating LLMs with CDMs through diffusion and fusion yields substantial gains in sparse-data educational contexts, offering a practical path toward more robust personalized learning systems.

Abstract

Cognitive Diagnosis (CD) has become a critical task in AI-empowered education, supporting personalized learning by accurately assessing students' cognitive states. However, traditional CD models often struggle in cold-start scenarios due to the lack of student-exercise interaction data. Recent NLP-based approaches leveraging pre-trained language models (PLMs) have shown promise by utilizing textual features but fail to fully bridge the gap between semantic understanding and cognitive profiling. In this work, we propose Language Models as Zeroshot Cognitive Diagnosis Learners (LMCD), a novel framework designed to handle cold-start challenges by harnessing large language models (LLMs). LMCD operates via two primary phases: (1) Knowledge Diffusion, where LLMs generate enriched contents of exercises and knowledge concepts (KCs), establishing stronger semantic links; and (2) Semantic-Cognitive Fusion, where LLMs employ causal attention mechanisms to integrate textual information and student cognitive states, creating comprehensive profiles for both students and exercises. These representations are efficiently trained with off-the-shelf CD models. Experiments on two real-world datasets demonstrate that LMCD significantly outperforms state-of-the-art methods in both exercise-cold and domain-cold settings. The code is publicly available at https://github.com/TAL-auroraX/LMCD

Figures (4)

• Figure 1: An illustration of the cold-start problem in cognitive diagnosis. (a) is hierarchical KC tree. (b) is sparse student-exercise interaction matrix. (c) is typical cognitive diagnosis framework for addressing cold-start problems.
• Figure 2: LMCD framework overview. (a) Knowledge Diffusion: LLMs generate enriched contents of exercises and knowledge concepts. (b) Semantic-Cognitive Fusion: Causal attention mechanisms integrate textual information with student-specific cognitive states to model relative difficulty.
• Figure 3: Impact of knowledge encoding strategies.
• Figure 4: Relative difficulty vs Absolute difficulty.