Cold Start Mitigation in an Adaptive Fact Learning System: Bayesian Prediction of Secondary School Student Performance

Bosch, Jelle (2021) Cold Start Mitigation in an Adaptive Fact Learning System: Bayesian Prediction of Secondary School Student Performance. Master's Thesis / Essay, Human-Machine Communication.

Preview	Text mHMC_2021_BoschJRA.pdf Download (485kB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (118kB)

Abstract

*Shortened*: Digital learning aids have become increasingly popular to support classical education. Particularly, adaptive fact learning systems (AFLSs) have been shown to lead to better learning outcomes as their underlying cognitive models allow them to test facts of appropriate difficulty at the appropriate time to optimise declarative memory reinforcement. With each given response, the cognitive model adapts to better reflect the user’s declarative memory, personalising the learning experience over time. When new, unknown facts or users are encountered, the cognitive model needs to adapt a number of times before it can provide an accurate representation, meanwhile the learning experience is suboptimal. This is known as the ‘cold start problem’. Presently, we sought to mitigate the cold start problem by predicting an AFLS parameter that reflected fact difficulty and student learning ability through five distinct Bayesian prediction methods. Predictions were made in a post-hoc simulation of a naturalistic dataset consisting of 117 million trials performed by 135 thousand Dutch secondary school students. Of the prediction methods, a fact-level prediction method that made predictions per fact based on previous performances on each fact and a hybrid prediction method that combined fact-level and student-level predictions were found able to mitigate the cold start problem.

Item Type:	Thesis (Master's Thesis / Essay)
Supervisor name:	Borst, J.P.
Degree programme:	Human-Machine Communication
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	20 Dec 2021 10:57
Last Modified:	20 Dec 2021 10:57
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/26367

Actions (login required)

View Item