Authors:
Noura Joudieh
1
;
Marwa Trabelsi
1
;
Ronan Champagnat
1
;
Mourad Rabah
1
and
Nikleia Eteokleous
2
Affiliations:
1
L3i Laboratory, La Rochelle University, La Rochelle, France
;
2
Frederick University, Cyprus
Keyword(s):
Learning Process, Trace Clustering, Process Mining, Learning Scenarios, Learning Paths, Quality of Education.
Abstract:
Learners adopt various learning patterns and behaviors while learning, rendering their experience a valuable asset for recommending learning paths for other learners. Process Mining is useful in this case to discover models that reveal learners’ taken learning paths in an educational platform. Nonetheless, due to the heterogeneity of behavior and the volume of data, trace clustering is crucial to reveal various groups of learners and discover relevant process models rather than ‘spaghetti’ ones. In this paper, we address the limits of and improve on a feature-based trace clustering approach known as FSS-encoding, ideal for unstructured processes to extract diverse learning patterns adopted by students, to be later employed in a learning path recommendation. Our enhancements include a refined pattern selection, preserving the uniqueness of less frequent events and increasing the overall effectiveness of the trace clustering process. Our method was applied to Moodle logs acquired from
2018 to 2022, comprising 471 students in the Computer Science and Engineering Department of Frederick University in Cyprus. The results show three clusters with a 25% improvement in silhouette coefficient. Their consequent discovered process models depict the various learning scenarios adopted, including activities like studying, solving exercises, undergoing assessments, applying, and others.
(More)