Capturing Inter-Individual Differences in Cognitive Processes via Hierarchical Clustering in Dynamic Bayesian Networks
Published in The Thirteenth Annual Conference on Advances in Cognitive Systems, 2026
Rebecca von Engelhardt, Malte Luttermann. (2026). "Capturing Inter-Individual Differences in Cognitive Processes via Hierarchical Clustering in Dynamic Bayesian Networks." Proceedings of the Thirteenth Annual Conference on Advances in Cognitive Systems. To appear.
Abstract
People systematically differ in how they integrate prior expectations and new evidence when forming beliefs about others. Yet they are not all different: individuals plausibly fall into groups that share similar cognitive patterns. Bayesian networks are a popular tool for modelling psychological processes, but a single network learned on the population level smooths away inter-individual differences, whereas fitting one network per individual yields noisy estimates that do not generalise. Clustering individuals into groups that share common patterns in belief revision, modelled by similar conditional probability tables, offers a principled compromise. However, choosing an appropriate distance metric to quantify similarity between individuals and a stopping criterion for the clustering process is non-trivial. In this paper, we transfer hierarchical agglomerative lifted model construction to the clustering of individuals in (dynamic) Bayesian networks. On simulated belief updating data, we study how central design choices of such studies affect the results, finding that the number of trials per individual matters substantially more than the number of participants. Overall, we demonstrate that hierarchical clustering recovers the underlying groups reliably while highlighting challenges that arise when transferring such mathematically developed clustering methods to the modelling of cognitive processes.