Enhancing Math Learning in an LMS Using AI-Driven Question Recommendations
Project Overview
The document explores the integration of generative AI in education, specifically focusing on enhancing math learning within a Learning Management System (LMS) through tailored question recommendations. Utilizing Meta's Llama-3.2-11B-Vision-Instruct model, the study evaluates three recommendation methodologies: cosine similarity, Self-Organizing Maps (SOM), and Gaussian Mixture Models (GMM). Findings indicate that cosine similarity yields accurate matches but risks producing repetitive recommendations, while SOM strikes an effective balance between similarity and novelty, resulting in improved user satisfaction. Conversely, GMM was found to be less effective due to subpar similarity computations. The research underscores the critical role of robust representation and suitable similarity metrics in the development of effective educational recommender systems, highlighting the potential of AI to enhance personalized learning experiences in mathematics. Overall, the document illustrates how generative AI can be leveraged to improve educational outcomes by providing customized content that meets individual learner needs.
Key Applications
AI-driven recommendation system for math questions
Context: Utilized in a Learning Management System (LMS) for SweSAT preparation, targeting students preparing for a standardized exam.
Implementation: Implemented using a combination of cosine similarity, Self-Organizing Maps (SOM), and Gaussian Mixture Models (GMM) based on user interaction data.
Outcomes: SOM yielded higher user satisfaction and engagement compared to cosine similarity and GMM, which underperformed.
Challenges: GMM faced challenges in effectively computing similarity, leading to lower user engagement.
Implementation Barriers
Technical Limitations
The GMM approach underperformed due to ineffective similarity computation using KL divergence.
Proposed Solutions: Future work could explore alternative metrics for similarity or revert to a clustering paradigm.
Project Team
Justus Råmunddal
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Justus Råmunddal
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