LIN, YU-CHING
Graduate Institute of Educational Information and Measurement, National Taichung University of Education PhD Student
E-mail：cms113107@gm.ntcu.edu.tw
WU, HUEY-MIN
Graduate Institute of Educational Information and Measurement, National Taichung University of Education Associate Professor
E-mail：whm@mail.ntcu.edu.tw
Abstract
With the rise of generative AI, constructing instructional systems that support both cognitive and affective needs is critical. This study investigates the effectiveness of Multi-Agent Systems (MAS) in a junior high school mathematics “Patterned Numbers” unit. A quasi-experimental pre-post design was employed, involving 36 ninth-graders in Taiwan divided into an experimental group (MAS) and a control group (Single-Agent System).
This study combined quantitative statistics with the frequency of behavioral coding of learners’ self-explanations. The results show that both groups of students made significant progress in mathematics achievement, and the multi-agent system model demonstrated greater remedial potential for learners with lower starting levels. In terms of behavioral patterns, the multi-agent AI-assisted system elicited cognitive behaviors such as reasoning, rationalization, and regulation, and was more effective than the single-agent AI-assisted system in guiding students toward deeper problem-solving thinking.
The findings confirm that a multi-agent AI-assisted system, through the division of labor among experts and a task-specific framework, can guide learners’ behavior from negative affective stagnation toward positive self-regulated learning. This study validated the effectiveness of multi-agent collaboration mechanisms in providing adaptive support in mathematics education and provided empirical evidence for the behavioral transformation mechanisms of AI-assisted learning. It offers significant reference value for the future development of personalized instructional systems equipped with affective support functions.
Keywords：Multi-Agent Systems（MAS）, Mathematics Education, Self-Regulated Learning（SRL）, Affective Learning, Behavioral Patterns