Integrating Performance and Side Factors into Embeddings for Deep Learning-Based Knowledge Tracing
Liangliang He · 2021
In computer-aided education systems, Deep Learning-based Knowledge Tracing (DLKT) models outperform traditional models on tracing learners’ knowledge in recent years. First, we propose a new Performance Factors-based Embedding (PFE) model for DLKT by extending learner’s historical performances on exercises into the existing Rasch Model-based Embedding (RME) model. Second, we find that side factors (e.g., template and hint) are helpful to reflect the individualized difficulties of different exercises covering the same skill by analysing data. Therefore, we introduce an extensible embedding framework to synthesize skill, exercise, performance factors and side factors, dubbed BPS. BPS consists of three components: base embedding, performance embedding and side embedding which allows one or more side factors related to the difficulty of exercise to be extended in BPS. Finally, experiments on three real-world benchmark datasets show that PFE and BPS significantly improve the state-of-the-art DLKT model on predicting future learner responses.