Uncertainty-aware Knowledge Tracing

2025-01-09

AI Summary

Existing knowledge tracing (KT) methods predominantly employ deterministic state representations, failing to capture the inherent uncertainty in students’ cognitive processes and thereby introducing estimation bias. To address this, we propose the first KT framework based on stochastic distributional embeddings: student knowledge states are modeled as probability distributions; a Wasserstein metric-guided self-attention mechanism captures the dynamic evolution of these distributions; and an aleatoric uncertainty-aware contrastive learning loss is introduced to enhance robustness. Evaluated on six real-world educational datasets, our method significantly outperforms state-of-the-art deep KT models—achieving new SOTA performance in both predictive accuracy and uncertainty robustness. By explicitly modeling cognitive uncertainty through distributional representations and principled geometric learning, our approach establishes a novel paradigm for interpretable and trustworthy intelligent educational assessment.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Intelligent Robots: State Estimation

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Knowledge Tracing (KT) is crucial in education assessment, which focuses on depicting students' learning states and assessing students' mastery of subjects. With the rise of modern online learning platforms, particularly massive open online courses (MOOCs), an abundance of interaction data has greatly advanced the development of the KT technology. Previous research commonly adopts deterministic representation to capture students' knowledge states, which neglects the uncertainty during student interactions and thus fails to model the true knowledge state in learning process. In light of this, we propose an Uncertainty-Aware Knowledge Tracing model (UKT) which employs stochastic distribution embeddings to represent the uncertainty in student interactions, with a Wasserstein self-attention mechanism designed to capture the transition of state distribution in student learning behaviors. Additionally, we introduce the aleatory uncertainty-aware contrastive learning loss, which strengthens the model's robustness towards different types of uncertainties. Extensive experiments on six real-world datasets demonstrate that UKT not only significantly surpasses existing deep learning-based models in KT prediction, but also shows unique advantages in handling the uncertainty of student interactions.

Problem

• Knowledge Tracing
• Uncertainty in Learning
• Student Learning State Assessment

Innovation

• Uncertainty Knowledge Tracing (UKT)
• Special Self-attention Mechanism
• Novel Learning Approach