Developing an AI learning companion for mathematics problem solving in elementary schools

Abstract

This study examined the impact of integrating the Taiwan Adaptive Learning Platform (TALP) with generative AI-based learning companion (TALPers) on fifth-grade students' mathematics learning performance. The integration was applied to both self-learning and remedial instruction, with both contexts incorporating Socratic dialogue and Pólya's problem-solving strategy. Results showed that video-based self-learning with TALPer support yielded the highest learning gains. In the remedial setting, students receiving TALPer support also demonstrated significantly better performance. Both approaches were particularly effective for low-achieving students. Lag Sequential Analysis (LSA) further revealed that high-achieving students engaged in more complex interaction patterns with TALPers, sustaining feedback cycles, whereas low-achieving students exhibited simpler interaction patterns. These findings provide empirical support for implementing AI-augmented learning systems in mathematics education and suggest that structured AI support can promote differentiated instruction and student engagement to help bridge achievement gaps.