Publicación: Exploring cognitive models to augment explainability in Deep Knowledge Tracing
Fecha
2023-06-13
Autores
Labra, Concha
Editor/a
Director/a
Tutor/a
Coordinador/a
Prologuista
Revisor/a
Ilustrador/a
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editor
ACM Digital Library
Resumen
Adaptive learning systems allow a personalized adaptation based on the characteristics of the student. Tracing the progress of knowledge and skills during the learning process through cognitive models is essential so that these systems can make appropriate decisions when carrying out personalization. This is the objective of Knowledge Tracing, which studies how to infer a cognitive model from the answers given to a sequence of questions or exercises. The incorporation of Deep Learning techniques in this field has given rise to Deep Knowledge Tracing (DKT) which usually has excellent predictive outcomes. The problem is that this increase in accuracy comes with a lack of explainability since Deep Learning models can be considered black boxes from which it is difficult to build interpretations or explanations. By contrast, traditional Knowledge Tracing methods are based on underlying learning models and provide a solid basis for explainability. In this paper we describe an ongoing research to build DKT models with a good trade-off between accuracy and explainability. To this end, we propose to use a loss function based on a mixup approach where the ground truth is a mix between the dataset labels and the predictions of a surrogate explainable model. The approach has potential to improve, not only explainability through the use of the surrogate, but also accuracy thanks to regularization effects. We will validate the approach by exploring, for different cognitive models, the trade-off curve that is obtained by plotting accuracy against explainability for different mixup values.
Descripción
Este artículo fue presentado en la 31 ACM Conference on User Modeling, Adaptation and Personalization (UMAP ’23 Adjunct), June 26–29, 2023, Limassol, Chipre.
This paper was presented at the 31st ACM Conference on User Modeling, Adaptation and Personalization (UMAP '23 Adjunct), June 26–29, 2023, Limassol, Cyprus.
Categorías UNESCO
Palabras clave
personalized learning systems, scrutable user models, explainability, deep knowledge tracing (DKT), cognitive models
Citación
Concha Labra and Olga C. Santos. 2023. Exploring cognitive models to augment explainability in Deep Knowledge Tracing. In UMAP ’23 Adjunct: Adjunct Proceedings of the 31st ACM Conference on User Modeling, Adaptation and Personalization (UMAP ’23 Adjunct), June 26–29, 2023, Limassol, Cyprus. ACM, New York, NY, USA, 4 pages. https://doi.org/10.1145/3563359.3597384
Centro
Facultades y escuelas::E.T.S. de Ingeniería Informática
Departamento
Inteligencia Artificial