Reproduce “From Correctness to Comprehension: AI Agents for Personalized Error Diagnosis in Education” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yi-Fan Zhang, Hang Li, Dingjie Song, Lichao Sun, Tianlong Xu, Qingsong Wen
Researchers introduce a multi-agent AI framework for personalized error diagnosis in education, shifting the focus from answer correctness to comprehensive, context-aware feedback. The work establishes MathCCS, a new multimodal benchmark with expert-annotated error types and real-world student data, demonstrating that integrating a time series agent with MLLMs improves error classification and feedback quality.
Key insights
- State-of-the-art MLLMs achieve low classification accuracy (below 30%) and generate poor quality suggestions (scoring 1-3 out of 10) on complex educational error diagnosis tasks.
- Leveraging students' historical learning data through a specialized Time Series Agent, particularly with modality-specific normalization and I-Conditioned/Teacher-force classifiers, significantly improves error classification robustness.
- A multi-agent approach, integrating the temporal analysis of a Time Series Agent with the generative capabilities of an MLLM Agent, substantially enhances both error classification and the quality of personalized feedback compared to MLLMs acting in isolation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.