Skip to main content Skip to navigation

Mathematical Word Problem Generation from Commonsense Knowledge Graph and Equations

Project Overview

The document highlights the innovative application of generative AI in education through the development of the MaKE model, designed to create mathematical word problems (MWPs) using commonsense knowledge and equations. By leveraging a combination of graph neural networks and conditional variational autoencoders, MaKE generates a diverse array of coherent MWPs, thereby alleviating the workload of teachers and enriching the learning experience for students. This approach facilitates varied practice exercises that go beyond rote memorization, promoting deeper understanding and engagement with mathematical concepts. Notably, the model demonstrates superior performance compared to existing methods, as evidenced by both automatic and human evaluation metrics, underscoring its potential to transform educational practices by providing personalized and adaptive learning resources. The findings indicate that generative AI can significantly enhance educational outcomes, making it a valuable tool in modern classrooms.

Key Applications

MaKE (Mathematical word problem generation from commonsense Knowledge and Equations)

Context: K-12 educational settings, targeting students learning mathematics

Implementation: An end-to-end neural model that generates MWPs by learning from graphs of symbolic equations and commonsense knowledge.

Outcomes: Provides diverse practice problems, reduces teacher workload, enhances student engagement and understanding of mathematical concepts.

Challenges: Generating MWPs that are both relevant and diverse, ensuring language coherence, and managing the complexity of mathematical relationships.

Implementation Barriers

Technical Barrier

Difficulty in generating MWPs that maintain both mathematical accuracy and natural language fluency.

Proposed Solutions: Utilizing gated graph neural networks for better understanding of mathematical relations and implementing a self-planning module to improve sentence structure.

Data Availability

Limited access to high-quality commonsense knowledge graphs for diverse MWP generation.

Proposed Solutions: Crowdsourcing verification of extracted commonsense knowledge and expanding the dataset through collaboration with educational institutions.

User Acceptance

Resistance from educators to adopt AI-generated MWPs due to concerns over quality and relevance.

Proposed Solutions: Conducting thorough evaluations and providing evidence of improved educational outcomes through the use of AI-generated content.

Project Team

Tianqiao Liu

Researcher

Qiang Fang

Researcher

Wenbiao Ding

Researcher

Hang Li

Researcher

Zhongqin Wu

Researcher

Zitao Liu

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Tianqiao Liu, Qiang Fang, Wenbiao Ding, Hang Li, Zhongqin Wu, Zitao Liu

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies