Original title: Structural Learning of BN2A models
Authors: Pérez Cabrera, Iván ; Vomlel, Jiří
Document type: Papers
Conference/Event: Workshop on Uncertainty Processing (WUPES’25), Třešť (CZ), 20250604
Year: 2025
Language: eng
Abstract: Bayesian networks are a popular framework for modeling probabilistic relationships between random variables and have been used successfully in educational tests. There is interest in a particular type of Bayesian networks we have called BN2A, which are characterized as bipartite networks, where the first layer consists of hidden variables (which commonly represent skills) and the second layer consists of observed variables (which represent questions in a test). In BN2A models all variables are assumed to be binary. The variables in the second layer depend on the variables in the first layer and this dependence is characterized by conditional probability tables representing Noisy-AND models. In this work, we propose an Expectation-Maximization (EM) algorithm for learning the structure of BN2A models, that is, for learning the relationship between hidden variables and observed variables. To test the structural learning algorithm, we conducted two experiments. For the first experiment, we used synthetic data generated from a BN2A model that we previously defined, while for the second experiment we used a well-known real-world dataset in the field of Cognitive Diagnostic Models, the Fraction Subtraction dataset. Our proposed algorithm has interesting potential use cases. One key application is to generate a reasonably accurate BN2A structure model for educational diagnosis, particularly in scenarios where no prior model exists. Depending on the required level of accuracy, the estimated model can be used directly to analyze skill profiles or serve as an initial framework for test designers, who can further refine it before implementation.
Keywords: Bayesian networks; Noisy-AND models; Psychometrical models
Host item entry: Proceedings of the 13th Workshop on Uncertainty Processing (WUPES’25), ISBN 978-80-7378-525-3

Institution: Institute of Information Theory and Automation AS ČR (web)
Document availability information: Fulltext is available at external website.
External URL: https://library.utia.cas.cz/separaty/2025/MTR/vomlel-0639050.pdf
Original record: https://hdl.handle.net/11104/0370050

Permalink: http://www.nusl.cz/ntk/nusl-692651


The record appears in these collections:
Research > Institutes ASCR > Institute of Information Theory and Automation
Conference materials > Papers