Paper recorded by Signals 4 on 2026-09-01 in cs.CL. Abstract reproduced from arXiv; link to the original below.
Published 2026-09-01 on arXiv · recorded by Signals 4 on 2026-09-02
Category: cs.CL · 自然语言处理 · first seen 2026-09-02
AI tutors are most useful when they adapt to each student's strengths, weaknesses, and preferred guidance, but evidence about which guidance works for which student is sparse, slow, and costly to collect from real learners. Student simulators can provide this signal as a proxy, yet existing approaches are limited: state-tracking models fit student behavior but struggle to process explanations or c