Knowledge tracing tasks involve analyzing students' historical question-answering sequences to assess their knowledge status, enabling the prediction of their future performance. Most existing knowledge tracing models only explore the direct impact of questions or knowledge concepts themselves on students' answers but do not effectively explore the deep relationships between questions and knowledge concepts, which weakens the predictive performance of the models. In addition, the nature of the deep learning black box leads to a lack of interpretability of the knowledge tracing model applying deep learning algorithms. Therefore, we introduce an interpretable deep knowledge tracing model with graph relationship information. It leverages graph attention neural networks to explore fine-grained relationships between questions and knowledge concepts in the original dataset, thus enhancing its predictive performance. Furthermore, DKVMN is employed to capture students' knowledge state for each knowledge concept, and IRT is incorporated to enhance interpretability. Our methodology beats popular knowledge tracing methods in terms of prediction performance, according to experimental results on three genuine public online education datasets, more accurately capturing students' knowledge states for better prediction of their future performance, and enhancing model interpretability.
Interpretable Deep Knowledge Tracing with Graph Relationship Information
11.10.2023
2660814 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Interpretable multi-hop knowledge reasoning for gastrointestinal disease
Springer Verlag | 2025
|IMGCN: interpretable masked graph convolution network for pedestrian trajectory prediction
Taylor & Francis Verlag | 2024
|DOAJ | 2020
|Weighted graph attention network-based interpretable vehicle trajectory prediction method and system
Europäisches Patentamt | 2024
|