Wenqian Xu, Feng Ji (2026) — RL-based measurement model for interactive assessments.
Key Findings
Interactive assessments generate sequential process data that conventional item response models (IRT) cannot adequately handle. This paper proposes a reinforcement learning measurement model that links action choices to state-action values, extending beyond existing MDP-based measurement approaches.^[raw/papers/2605.09305.md]
The model addresses the gap between traditional static assessment models and the dynamic, interactive nature of modern computer-based assessments. It builds on prior work (LaMar, 2018) but improves reliability of estimates for interactive assessments where students' action sequences carry diagnostic information.^[raw/papers/2605.09305.md]
Connections to AIED
This work directly extends knowledge-tracing-irt by replacing static IRT with a dynamic RL-based approach. It connects to assessment by providing a new framework for scoring interactive assessments that capture student problem-solving processes rather than just final answers.^[raw/papers/2605.09305.md]
The approach also relates to knowledge-tracing and learning-analytics as it extracts diagnostic signals from interaction data that go beyond traditional performance metrics.
Related Pages
- knowledge-tracing-irt — RL measurement model extends beyond static IRT approaches
- assessment — New framework for scoring interactive process-based assessments
- knowledge-tracing — Dynamic tracing of student performance through sequential action data
- learning-analytics — Extracting diagnostic signals from interactive assessment logs
- intelligent-tutoring — Interactive assessment models inform tutoring system design
- representation-robustness-llm-math-problem-solving — Representation Robustness under Executable Reasoning Constra
Citation
APA: Xu, W., & Ji, F. (2026). Reinforcement learning measurement model. arXiv:2605.09305.