New embeddings prevent label leakage in knowledge tracing models

Researchers have developed a novel approach to enhance knowledge tracing models by preventing label leakage and incorporating recency encoding. The proposed method masks ground-truth labels during input embedding construction, similar to BERT's masked language modeling, to avoid inadvertent answer revelation. Additionally, a new Recency Encoding technique captures the temporal distance between interactions, better modeling learning dynamics like forgetting. These embeddings have shown consistent improvements in prediction accuracy when integrated into existing knowledge tracing models such as DKT, DKT+, AKT, and SAKT across various benchmarks. AI

IMPACT This research could lead to more accurate educational tools by improving how student learning is modeled and predicted.

RANK_REASON The cluster contains a research paper detailing a new method for knowledge tracing models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• BERT
• Deep Knowledge Tracing
• DKT+
• Recency Encoding
• Yahya Badran

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →