Do we need to go Deep? Knowledge Tracing with Big Data
Project Overview
The document examines the role of Interactive Educational Systems (IES) in enhancing knowledge tracing and modeling learner performance, emphasizing the comparative efficacy of traditional logistic regression models versus deep learning approaches in predicting student outcomes. Utilizing extensive datasets, particularly the EdNet dataset, the findings reveal that traditional models, which rely on carefully engineered features, frequently yield better performance than deep learning models, particularly across varying dataset sizes. This suggests that while deep learning has potential, its effectiveness in educational contexts remains inconclusive and necessitates further empirical validation. Furthermore, the document underscores the critical importance of model interpretability in educational AI, advocating for a balanced approach that combines the strengths of both traditional and modern techniques to optimize student learning outcomes. Overall, it calls for ongoing research to fully understand the practicality and impact of generative AI in education.
Key Applications
Interactive Educational Systems (IES)
Context: Educational settings, particularly in developing countries lacking qualified tutors, targeting students learning skills such as English and mathematics.
Implementation: Implemented by utilizing student interaction data from IES systems to develop learner performance models.
Outcomes: Provides personalized learning paths and feedback comparable to human tutoring, while being cost-effective.
Challenges: Ensuring accuracy and interpretability of performance models, balancing the trade-off between model complexity and user understanding.
Implementation Barriers
Technical Barrier
The performance of deep learning models in education is not yet satisfactory compared to traditional models.
Proposed Solutions: Further examination of the relationship between modeling techniques and dataset characteristics is needed. Incorporating advancements in data science methodologies can enhance model performance.
Project Team
Varun Mandalapu
Researcher
Jiaqi Gong
Researcher
Lujie Chen
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Varun Mandalapu, Jiaqi Gong, Lujie Chen
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