TheoremExplainAgent: Explaining Math Formulas and Scientific Theorems via Animated Videos 194

What is TheoremExplainAgent?

TheoremExplainAgent (TEA) is a STEM-domain theorem visualization and explanation system based on a dual-agent architecture. It is designed to intuitively explain mathematical formulas and scientific theorems by generating animated videos.

The Technical Architecture of TheoremExplainAgent

• Dual-Agent Collaboration Framework: TheoremExplainAgent adopts a dual-agent collaboration workflow consisting of a Planner Agent and a Coding Agent. The Planner Agent is responsible for theorem parsing and instructional planning, while the Coding Agent transforms the lesson plans into Manim animation code.

• Core Technical Support: It uses Manim as the animation engine, combines RAG (Retrieval-Augmented Generation) technology to integrate disciplinary knowledge bases, and employs TheoremExplainBench (TEB) as its evaluation system.

Core Components of TheoremExplainAgent

• Planner Agent: Responsible for understanding theorem content, formulating teaching plans, designing lesson plans, and writing explanation scripts.

• Coding Agent: Converts the lesson plans generated by the Planner Agent into Manim animation code, produces 5-minute animated videos, and pairs them with corresponding audio narration.

Technical Highlights of TheoremExplainAgent

• Multimodal Output Capability: Supports mixed presentations of animations, graphics, and formulas, making teaching content more vivid and engaging.

• Rendering Speed Improvement: Rendering speed is increased by 20% compared with traditional methods.

• Innovative Evaluation System: Pioneers a five-dimensional quantitative evaluation standard, with a Spearman correlation coefficient of 0.88.

• Lightweight Deployment: The o3-mini model reduces parameters by 40% while maintaining 93.8% of its core performance.

Application Scenarios of TheoremExplainAgent

• MOOCs Platforms: Such as Coursera and Khan Academy, used for course development and automatically generating teaching animation materials.

• Educational Resource Repositories: Such as OpenStax and LibreTexts, used for content enhancement and automatic generation of dynamic textbooks.

• Classroom Teaching Assistance: Used as supplementary material in traditional classrooms to showcase key knowledge points.

• Self-Study Tool: Students can watch these videos after class to review and reinforce what they have learned.

• Online Education Platforms: Platforms can use this tool to enrich the multimedia resources in their course libraries.

• Home Tutoring: Parents can use these videos to provide additional help and support for their children.

Technical Advantages of TheoremExplainAgent

• Improved Manual Evaluation Efficiency: Manual evaluation efficiency is increased by 4 times compared with traditional methods.

• LaTeX Formula Parsing Accuracy: As high as 98.6%.

• Support for Disciplinary Symbol Systems: Supports 20+ disciplinary symbol systems, with compatibility for multi-model APIs (GPT, Claude, Gemini).

Project Address:https://tiger-ai-lab.github.io/TheoremExplainAgent/

Paper Address:https://arxiv.org/pdf/2502.19400