Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·Aug 11·5 min read
Enhancing Knowledge Tracing through Leakage-Free and Recency-Aware Embeddings
TL;DR
A new approach to Knowledge Tracing (KT) enhances prediction accuracy by preventing label leakage and incorporating recency-aware embeddings. This method improves performance across various KT models and benchmarks.
✦ Why It Matters
Implement masking techniques and recency encoding in your KT models to enhance prediction accuracy today.
Key Takeaways
- Use a MASK label to prevent label leakage in KT models.
- Incorporate Recency Encoding to model forgetting in learning.
- Test improvements on existing models like DKT and AKT.
- Evaluate performance across multiple KT benchmarks for validation.
How It Works
The proposed method introduces a MASK label to replace ground-truth labels during input embedding construction, preventing label leakage. Recency Encoding is implemented to track the time since a student's last interaction with an item, allowing the model to better account for forgetting, a key aspect of human learning.