Skip to main content Skip to navigation

A Benchmark for Math Misconceptions: Bridging Gaps in Middle School Algebra with AI-Supported Instruction

Project Overview

The document explores the incorporation of generative AI, specifically Large Language Models (LLMs), in middle school algebra education to identify and address student misconceptions. It introduces a dataset comprising 55 algebra-related misconceptions along with diagnostic examples to aid educators in recognizing and remediating math errors. The research assesses the performance of GPT-4 in detecting these misconceptions, revealing its potential to enhance personalized learning experiences for students. However, the findings also underscore challenges in accuracy, particularly concerning complex topics such as ratios and proportional reasoning. Feedback from educators suggests a significant interest in leveraging AI tools for diagnostic purposes, indicating the promise of generative AI in supporting teachers and improving student understanding in mathematics. Overall, while generative AI shows potential in transforming educational practices, careful consideration of its limitations is necessary to maximize its effectiveness in addressing student learning needs.

Key Applications

Benchmark for Algebra Misconceptions using AI-supported instruction

Context: Middle school algebra education targeting low-income and minoritized students

Implementation: Development of a dataset comprising 55 misconceptions and 220 diagnostic examples, with evaluation of GPT-4's performance in identifying these misconceptions

Outcomes: Achieved 83.9% accuracy in diagnosing misconceptions when constrained by topic and utilizing educator feedback; educators expressed interest in using the dataset for diagnosis

Challenges: Struggled with identifying specific types of misconceptions, especially in ratios and proportional reasoning; varied familiarity with AI tools among educators

Implementation Barriers

Technological barrier

Challenges in accurately identifying specific math misconceptions using AI models

Proposed Solutions: Incorporating topic-constrained testing and educator feedback to improve AI diagnostic capabilities; exploring multimodal approaches for better engagement with complex topics

Educational barrier

Limited familiarity of educators with AI tools and their application in diagnosing student misconceptions

Proposed Solutions: Providing training and resources for educators on using AI tools effectively in the classroom

Project Team

Otero Nancy

Researcher

Druga Stefania

Researcher

Lan Andrew

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Otero Nancy, Druga Stefania, Lan Andrew

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