Reproduce “Cold Start Problem: An Experimental Study of Knowledge Tracing Models with New Students” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Indronil Bhattacharjee, Christabel Wayllace
Researchers at New Mexico State University investigated the "cold start problem" for new students in Knowledge Tracing (KT) models by evaluating Deep Knowledge Tracing (DKT), Dynamic Key-Value Memory Networks (DKVMN), and Self-Attentive Knowledge Tracing (SAKT) using a novel framework. The study demonstrates that all models initially struggle but progressively improve with more interactions, with DKVMN showing rapid early adaptation and DKT exhibiting robust long-term learning.
Key insights
- The cold start problem for new students poses a severe challenge for even state-of-the-art deep learning KT models, highlighting the critical need for improved generalization capabilities.
- Model architectures demonstrate varying strengths in cold start scenarios: DKVMN offers rapid initial adaptation, DKT excels in sustained long-term learning, and SAKT provides quick early gains, though it may plateau.
- This study establishes a more realistic benchmark for future KT research, emphasizing the development of hybrid models, transfer learning, or few-shot/zero-shot learning techniques to enable immediate and effective personalization for new users.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.