AI-Driven Virtual Teacher for Enhanced Educational Efficiency: Leveraging Large Pretrain Models for Autonomous Error Analysis and Correction

TL;DR

The paper tackles the high cost and limited generalizability of traditional math error correction by proposing VATE, a multimodal, dual-stream architecture that uses student drafts, an error pool, and guided dialogue to autonomously identify and remediate errors. It introduces a multimodal data pipeline and two error-analysis streams to achieve open-world error detection with reduced reliance on exhaustive error databases. Empirical results from deployment on the Squirrel AI platform show 78.3% expert-validated error analyses, substantial learning gains (e.g., 15% reduction in related error rates, up to 37% with effective dialogue, NVRS down 61.5%), and broad scalability across regions. The work demonstrates that AI-driven, draft-centric error analysis can transform educational efficiency, offering a cost-effective, flexible solution adaptable to multiple subjects and grade levels.

Abstract

Students frequently make mistakes while solving mathematical problems, and traditional error correction methods are both time-consuming and labor-intensive. This paper introduces an innovative \textbf{V}irtual \textbf{A}I \textbf{T}eacher system designed to autonomously analyze and correct student \textbf{E}rrors (VATE). Leveraging advanced large language models (LLMs), the system uses student drafts as a primary source for error analysis, which enhances understanding of the student's learning process. It incorporates sophisticated prompt engineering and maintains an error pool to reduce computational overhead. The AI-driven system also features a real-time dialogue component for efficient student interaction. Our approach demonstrates significant advantages over traditional and machine learning-based error correction methods, including reduced educational costs, high scalability, and superior generalizability. The system has been deployed on the Squirrel AI learning platform for elementary mathematics education, where it achieves 78.3\% accuracy in error analysis and shows a marked improvement in student learning efficiency. Satisfaction surveys indicate a strong positive reception, highlighting the system's potential to transform educational practices.

Figures (8)

• Figure 1: The operational mode of our Virtual AI Teacher system: After assigning problems to students, it accepts the students' drafts and answers. First, it compares them with our internal error cause database; if a matching error is found, it directly provides the error analysis. If not, it utilizes our large model system for error analysis. The system then engages in multiple rounds of dialogue with the student, guiding them to correct the corresponding errors based on the error analysis in each conversation.
• Figure 2: Architecture of VATE. The VATE framework is a multi-agent system that begins by analyzing a user-inputted draft image using our specialized Draft Prompt mechanism. This draft analysis, alongside student responses, problem explanations, and correct answers, is then formatted according to our predefined structure and inputted into another language model. The system ultimately produces a detailed error analysis and recommendations, guiding students on how to correctly approach and solve similar problems.
• Figure 3: Distribution of the number of different incorrect answers across 2,326 questions.
• Figure 4: The VATE system deployed in the Squirrel Ai Learning machine interface, displaying the problem statement, solution explanation, student's draft, and AI dialogue options.
• Figure 5: VATE Learning Summary Pages: (left): Overview of the student's learning content for the day, including study duration and associated knowledge points. (middle) Detailed analysis of the student's mastery across various relevant knowledge points. (right) Comparative analysis of the student's problem-solving time versus the national average.