A Study on Personalized Programming Learning for High School Students Supported by Large Language Models

Abstract

This study investigates the impact of an LLM-based AI
teaching assistant on personalized programming learning for
high school students and examines its effects on learning
achievement, self-efficacy, learning attitudes, and AI
acceptance. Using a single-group pretest-posttest design,
thirty students from an information technology course at a
senior high school in Hsinchu, Taiwan participated in a
twenty-week experiment. The first ten weeks involved
traditional instruction; the subsequent ten weeks
introduced an LLM-based AI teaching assistant providing
personalized explanations, code diagnosis, error feedback,
and conceptual guidance based on individual proficiency
levels. Research instruments included a programming
achievement test, self-efficacy scale, learning attitude
scale, and AI acceptance scale. Quantitative data were
analyzed using descriptive statistics, paired-sample
t-tests, and Wilcoxon signed-rank tests. Qualitative data
from student interviews, AI interaction logs, and teacher
observations were analyzed through thematic analysis and
cross-validation.