Search and Filter

Pats: Personality-Aware Teaching Strategies With Large Language Model Tutors

Date

AI Overview

Why it matters

Mismatches between tutoring strategies and student personalities can harm learning, yet current AI tutoring systems rely on generic question-answering without tailoring their instructional strategies to student traits.

Study at a glance
  • Question: Can AI tutors personalize their teaching strategies to better support learners with diverse personality profiles?
  • Setting and data: Italy and online; 110 high school students' chat logs, 4 expert teachers, 120 Prolific teachers, and simulated dialogues across 32 personality profiles.
  • Method: System development and simulated dialogue evaluation with expert teacher and LLM judges.
  • AI tools: PATS, GPT-4o, Llama V3.3 70B, Gemini-2.0-Flash, ChatGPT Pro
Key findings
  • Human teachers consistently preferred the personality-aware tutoring framework over both simple and personality-informed baselines across all evaluated dimensions.
  • Three evaluated large language models significantly favored the personality-aware framework over baselines across story and image description tasks.
  • Annotators found that the framework successfully implemented an intended personality-aligned teaching strategy in 94.6% of sampled dialogues.

Summary drafted with AI from the full paper and may contain errors. See the full paper for details.

Recent advances in large language models (LLMs) demonstrate their potential as educational tutors. However, different tutoring strategies benefit different student personalities, and mismatches can be counterproductive to student outcomes. Despite this, current LLM tutoring systems do not take into account student personality traits. To address this problem, we first construct a taxonomy that links pedagogical methods to personality profiles, based on pedagogical literature. We simulate student-teacher conversations and use our framework to let the LLM tutor adjust its strategy to the simulated student personality. We evaluate the scenario with human teachers and find that they consistently prefer our approach over two baselines. Our method also increases the use of less common, high-impact strategies such as role-playing, which human and LLM annotators prefer significantly. Our findings pave the way for developing more personalized and effective LLM use in educational applications.