No Task Left Behind: Multi-Task Learning of Knowledge Tracing and Option Tracing for Better Student Assessment
Project Overview
The document explores the innovative application of generative AI in education through the introduction of a multi-task learning framework known as Dichotomous-Polytomous Multi-Task Learning (DP-MTL), designed specifically for student assessment in AI Education. By integrating Knowledge Tracing (KT) and Option Tracing (OT), DP-MTL aims to enhance the precision of student evaluations by accounting for both the correctness of responses and the specific options selected by students. The authors provide evidence that this framework significantly boosts performance in tasks related to knowledge state tracking and score prediction. As a result, DP-MTL represents a notable advancement in personalized education technologies, offering the potential to tailor learning experiences to individual student needs and thereby improve educational outcomes.
Key Applications
Dichotomous-Polytomous Multi-Task Learning (DP-MTL)
Context: Student assessment in educational environments, targeting students undergoing assessments through multiple choice questions.
Implementation: DP-MTL combines KT and OT using a multi-task learning framework that incorporates both correctness and option choice predictions.
Outcomes: Significant improvements in the accuracy of knowledge tracing, option tracing, and score prediction performances.
Challenges: Complexity in integrating both KT and OT seamlessly and ensuring the model handles various data sparsity effectively.
Implementation Barriers
Technical
Challenges in effectively combining the KT and OT tasks within a multi-task learning framework, as well as issues related to model performance due to data sparsity.
Proposed Solutions: The proposed architecture design for DP-MTL aims to facilitate integration and improve model performance through regularization techniques. The methodology includes strategies for handling different levels of data sparsity through hyper-parameter tuning and adjusting the mixing ratio in the DP-MTL framework.
Project Team
Suyeong An
Researcher
Junghoon Kim
Researcher
Minsam Kim
Researcher
Juneyoung Park
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Suyeong An, Junghoon Kim, Minsam Kim, Juneyoung Park
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