FEAT: A Preference Feedback Dataset through a Cost-Effective Auto-Generation and Labeling Framework for English AI Tutoring
Project Overview
The document explores the application of a generative AI framework known as FEAT, which is designed to automate teacher feedback in English tutoring, highlighting the critical role of feedback in the learning process. It details the creation of three datasets—DIRECT-Manual, DIRECT-Generated, and DIRECT-Augmented—and demonstrates that even a limited amount of high-quality human-annotated feedback can substantially boost the effectiveness of AI models. The findings indicate that generative AI not only streamlines the feedback generation process, significantly cutting down on time and costs, but also enhances educational outcomes for students. Overall, the implementation of generative AI in education, particularly through automated feedback mechanisms, shows promise in improving learning experiences and efficiency.
Key Applications
FEAT - Feedback Dataset Generation Framework
Context: English language tutoring for students
Implementation: Developed a cost-effective framework using LLMs to generate teacher feedback and constructed three datasets for training AI models.
Outcomes: Demonstrated improved performance of AI models with minimal human annotation required, enhancing the scalability of teacher feedback.
Challenges: High-quality human feedback is time-consuming and costly to generate; the AI-generated feedback may not always meet educational standards.
Implementation Barriers
Cost Barrier
Generating high-quality teacher feedback data is time-consuming and expensive.
Proposed Solutions: Use of generative AI frameworks to automate and reduce the cost of feedback generation.
Quality Barrier
AI-generated feedback may lack the quality and specificity of human feedback. To enhance quality, a small portion of human-annotated feedback can be incorporated into AI-generated datasets.
Proposed Solutions: Incorporating human-annotated feedback into AI-generated datasets.
Project Team
Hyein Seo
Researcher
Taewook Hwang
Researcher
Yohan Lee
Researcher
sangkeun Jung
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Hyein Seo, Taewook Hwang, Yohan Lee, sangkeun Jung
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