Cold Start Problem: An Experimental Study of Knowledge Tracing Models with New Students
Project Overview
The document examines the challenges faced by Knowledge Tracing (KT) models in Intelligent Tutoring Systems (ITS), particularly focusing on the cold start problem, which affects the ability to predict knowledge states for new students with limited prior interaction data. It evaluates three prominent KT models—Deep Knowledge Tracing (DKT), Dynamic Key-Value Memory Networks (DKVMN), and Self-Attentive Knowledge Tracing (SAKT)—across various datasets. The research reveals that while all models demonstrate improvement with increased student interactions, their initial performances vary significantly, especially in cold start situations. The findings underscore the necessity for enhancing these models to better generalize to new learners and highlight the importance of developing more robust models capable of effective few-shot and zero-shot learning. Overall, the document emphasizes the potential of generative AI in addressing key challenges within educational technology, particularly in personalizing learning experiences and improving educational outcomes through better knowledge state predictions.
Key Applications
Knowledge Tracing models (DKT, DKVMN, SAKT)
Context: Educational context focusing on Intelligent Tutoring Systems aimed at new students with limited interaction data.
Implementation: Models trained on historical data from past students and tested on entirely new students to assess their cold start performance.
Outcomes: All models show improved accuracy with more interactions; DKVMN performs best initially, while DKT improves steadily over time.
Challenges: All models struggle initially under cold start conditions, indicating limitations in predicting knowledge states for new students.
Implementation Barriers
Technical
Cold start problem where models struggle to accurately predict new students' knowledge states due to minimal initial data.
Proposed Solutions: Developing hybrid KT models that enhance existing approaches and exploring advanced machine learning techniques like transfer learning.
Project Team
Indronil Bhattacharjee
Researcher
Christabel Wayllace
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Indronil Bhattacharjee, Christabel Wayllace
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