The Imperfect Learner: Incorporating Developmental Trajectories in Memory-based Student Simulation

TL;DR

This paper tackles the challenge of simulating imperfect, developing learners by introducing SimLearner, a memory-based framework that models developmental trajectories through a three-level hierarchical memory system comprising episodic memory $M_e$, conceptual memory $M_c$, and metacognitive profiling $M_s$, all embedded within a curriculum-aligned NGSS knowledge structure. It combines structured knowledge representation with dynamic forgetting and consolidation dynamics, and augments learner profiles with personality traits and metacognitive skills to reflect individual differences. The NGSS-based curriculum hierarchy and the memory consolidation mechanisms enable simulators to exhibit grade-appropriate knowledge trajectories across grades 1–5, while restricting access to advanced concepts to maintain realism. Experiments show that SimLearner yields curriculum-aligned concept mastery, consistent grade-level progression, and interpretable effects of personality and metacognition on learning, offering a scalable tool for evaluating educational AI systems without real participants. The framework can be extended to diverse domains and supports rigorous analysis of tutoring strategies and cognitive development in AI-driven educational contexts.

Abstract

User simulation is important for developing and evaluating human-centered AI, yet current student simulation in educational applications has significant limitations. Existing approaches focus on single learning experiences and do not account for students' gradual knowledge construction and evolving skill sets. Moreover, large language models are optimized to produce direct and accurate responses, making it challenging to represent the incomplete understanding and developmental constraints that characterize real learners. In this paper, we introduce a novel framework for memory-based student simulation that incorporates developmental trajectories through a hierarchical memory mechanism with structured knowledge representation. The framework also integrates metacognitive processes and personality traits to enrich the individual learner profiling, through dynamical consolidation of both cognitive development and personal learning characteristics. In practice, we implement a curriculum-aligned simulator grounded on the Next Generation Science Standards. Experimental results show that our approach can effectively reflect the gradual nature of knowledge development and the characteristic difficulties students face, providing a more accurate representation of learning processes.

Figures (8)

• Figure 1: One example of memory-based student simulation: When the knowledge setting is on basic water concepts and the student is asked a harder question, SimLearner demonstrates uncertainty and guessing behavior (as highlighted), while the base method directly provides the correct answer.
• Figure 2: Overview of student simulation architecture built on hierarchical memory mechanism, structured knowledge representation, and multi-dimensional profiling (cognitive, noncognitive, metacognitive aspects).
• Figure 3: The structured knowledge representation grounded on the NGSS curriculum standard.
• Figure 4: Results of knowledge mastery simulation as a grade-1 student. Lower scores in grades 2-5 indicate better alignment with natural developmental levels.
• Figure 5: Comparison of grade-level knowledge coverage according to the curriculum standard from grade 1 to 5. Compared with the base method (red line), SimLearner shows a much smaller gap with the expected coverage.