Predicting Learning Performance with Large Language Models: A Study in Adult Literacy
Project Overview
The document examines the role of Large Language Models (LLMs), specifically GPT-4, in enhancing adult literacy education by predicting learning performance through Intelligent Tutoring Systems (ITSs). It emphasizes the synergy between LLMs and traditional machine learning models, which improves the accuracy of predictions and personalizes the educational experience for learners. By analyzing performance data, LLMs can refine instructional strategies and effectively identify at-risk individuals, thereby fostering more effective adult literacy training programs. The findings underscore the transformative potential of generative AI in education, showcasing its ability to tailor learning experiences and optimize educational outcomes for adult learners.
Key Applications
Intelligent Tutoring Systems (ITSs) using GPT-4 for learning performance prediction
Context: Adult literacy education programs, particularly reading comprehension
Implementation: Utilization of reading comprehension datasets from AutoTutor to evaluate GPT-4's predictive capabilities against traditional machine learning methods like Bayesian Knowledge Tracing and XGBoost.
Outcomes: Enhanced predictive accuracy for learner performance, improved personalization of literacy instruction, and identification of at-risk learners.
Challenges: Stability issues with automated model tuning and variability in prediction performance.
Implementation Barriers
Technical Barrier
Constraints related to the fine-tuning of LLMs may limit the optimization of models specific to the dataset. Limitations in executing deep learning models restrict the application of more advanced techniques such as Deep Knowledge Tracing.
Proposed Solutions: Continued exploration of the integration of LLMs with traditional machine learning models to improve predictive capabilities. Future research could explore alternative architectures or methodologies to enhance predictive accuracy.
Project Team
Liang Zhang
Researcher
Jionghao Lin
Researcher
Conrad Borchers
Researcher
John Sabatini
Researcher
John Hollander
Researcher
Meng Cao
Researcher
Xiangen Hu
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Liang Zhang, Jionghao Lin, Conrad Borchers, John Sabatini, John Hollander, Meng Cao, Xiangen Hu
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