Design and evaluation of ChatGPT-MWPS: an AI-enhanced learning system for improving primary students’ mathematical word problem solving

Abstract Mathematical word problem solving is a fundamental aspect of primary mathematics education, yet students often struggle with translating textual descriptions into structured mathematical representations. In response to this challenge, this study introduces and evaluates ChatGPT-MWPS, an AI-enhanced learning system designed to support fifth-grade students in developing mathematical problem-solving skills through adaptive scaffolding and real-time, step-by-step feedback. Built on a Java-based framework and incorporating OpenAI’s generative AI technology, the system interprets problem content and delivers personalized solution guidance within a digital test paper environment. It supports key instructional functions, such as batch question import, automated test generation, and user management, enabling flexible application in both classroom and self-directed learning settings. A quasi-experimental design was employed with 52 fifth-grade participants to assess the system’s usability and potential educational effectiveness. System perofrmance was measured using a standardized usability scale with 13 dimensions . Results demonstrate that ChatGPT-MWPS achieved favorable usability ratings across the measured dimensions, with particularly strong performance in system reliability, information quality, and cognitive engagement. These findings establish the system's potential for enhancing mathematical learning experiences through its well-received usability characteristics and functional effectiveness. This research contributes to the growing domain of AI in education by demonstrating how generative AI can facilitate personalized mathematics instruction in primary school settings. The outcomes provide practical implications for the design of intelligent tutoring systems that aim to enhance mathematical reasoning and promote learner autonomy.