Knowledge tracing for adaptive learning in a metacognitive tutor

Abstract

Adaptive learning is provided in intelligent tutoring systems (ITS) to enable learners with varying abilities to meet their expected learning outcomes. Despite the personalized learning afforded by ITSes using adaptive learning, learners are still susceptible to shallow learning. Introducing metacognitive tutoring to teach learners how to be aware of their knowledge can enable deeper learning. However, metacognitive tutoring on top of cognitive tutoring can lead to unsustainable cognitive loads. Using metacognitive inputs for knowledge tracing was explored for managing cognitive loads. Hidden Markov models (HMM) and artificial neural networks were used to train models on a synthetic dataset created from predetermined learner personas. The models created with metacognitive inputs were compared with the models created without said inputs. The models using metacognitive inputs performed better than the standard models while still following learning intuitions. This indicates that combining knowledge tracing and metacognitive tutoring is a viable option for improving learning outcomes. This is an important finding since online learning, which demands metacognitive skills, is becoming popular for various topics, including those that are challenging even with immediate teacher assistance.

Reflection on the Construction and Impact of an Adaptive Learning Ecosystem

“…Adaptive metacognitive assistance, cognitive maps and intervention models such as adaptive navigators, mind mapping, and personalized learning analytics dashboards are provided to allow self-regulated learning (Carlon & Cross, 2022;Lim et al, 2023). The validation, development and innovation of adaptive learning with personal traits requires theoretical and empirical assessments based on the evidence and recordings of an individual lifelong learning cycle.…”

Adaptive Knowledge Control in Digital Learning as a Factor in Improving the Quality of Education

“…To describe the adaptive testing model, appropriate mathematical models can be used, which are also used to assess the probabilities of the ability level of the test subject and provide a choice of difficulty for the next task during testing. One such model is a Markov random process with discrete states and continuous and discrete time [16].…”

“…The basic concept of adaptive testing is the consistent presentation of tasks to the test subject, the level of complexity of which is determined by networks or Markov chains [16]. In the case of solving the problem by the test subject, when being in x k a state, a transition occurs to the next, x k+1 , state, otherwise-to the previous, x k−1 , state.…”

“…The relationship between the levels of complexity of the test task and the readiness of the test subject can be described by a model based on fuzzy mathematics [12], logical models of Rush and Birnbaum [13], models using Bayesian networks [14,15], Markov networks [16], Petri nets [17], and finite automata [18].…”

An Introduction to Bayesian Knowledge Tracing with pyBKT

“…Among the widely recognized learner models, knowledge tracing particularly stands out based on its ability to trace learners' knowledge and predict their knowledge states [6][7][8]. Within adaptive learning systems, it is essential to construct a knowledge-tracing model for all learners interacting with the system, representing their proficiency levels in all skills taught in the course [9,10]. For example, an online learning platform offering training in various language skills (e.g., vocabulary, grammar, and conversation) can use a knowledge tracing model to track each learner's progress and proficiency.…”