Traditional Knowledge Tracing Models for Clustered Students
Deliang Wang, Zhang Zhi, Jiachen Song, Yujie Lu · The Educational Review USA · 2021
Against the background of the worldwide COVID-19 pandemic, online learning has currently become one of the dominant educational forms.For more effective online learning, knowledge tracing that can dynamically estimate the knowledge state of learners should be paid more attention.This study gives an overall introduction of knowledge tracing models.We illustrate the development of the Bayesian Knowledge Tracing model and clarify its concrete mathematic principle.In addition, an individualized method based on clustered students for the Bayesian Knowledge Tracing model is initially proposed, which changes the individualization level from a group of all students to subgroups.To confirm whether this individualized method can be generalized to other knowledge tracing models, we also test it on logistic knowledge tracing models.Therefore, we provide an introduction about the principles of three logistic knowledge tracing models.We evaluate our method on the four models with two international public educational datasets.The results show that all knowledge tracing models for clustered students outperform the original models without clustering.The Bayesian Knowledge Tracing model and item response theory for clustered students both gain good improvement in their performance, while the Performance Factors Analysis for clustered students has only few improvements.The reason for the experimental results is discussed at the end.Overall, this paper provides a general method for further promoting the performance of traditional knowledge tracing models that requires more studies.