A Framework for Dynamic User Modeling Integrating Data Stream Mining and Process Mining in Educational Contexts

Abstract

A user model (UM) is a representation of the characteristics, behaviors, and preferences of a user that a computer system utilizes to provide the user with personalized experiences. In educational settings, for instance, a system can use information about the Learning Styles (LSs) of a student to provide them with adequate study materials. Furthermore, with the growing use of Learning Management System (LMS) platforms, such as Moodle, continuous streams of user data highlight the need to compute student models dynamically to support adaptive decision-making in teaching and learning. This paper introduces a framework for Dynamic User Modeling (DUMING) to create evolving Dynamic User Models (DUMs) that capture and adapt to changes in student LSs over time, based on event log streams from an LMS. This framework leverages the Felder and Silverman Learning Styles Model (FSLSM) and integrates Data Stream Mining (DSM) and Process Mining (PM) techniques across three layers: collection, preprocessing, and mining. Collected event streams are processed within landmark windows and classified into High-Level Events (HLEs) based on LMS functionality. The STREAMKMeans algorithm clusters events, and the PM phase uses Alpha Miner (AM), Heuristic Miner (HM), and Inductive Miner (IM) algorithms to identify learning paths represented as Petri Nets (PNs), with IM showing the highest values in the quality metrics used. Implemented and evaluated in a course using Moodle at the Universidad Veracruzana in Mexico, the framework employs IF-THEN rules to relate HLE proportions to FSLSM-based LSs, visualizing changes through radar charts. The results demonstrate the effectiveness of the framework in generating adaptive DUMs, providing insights that enhance over time educational support and enable timely interventions.