Integrating SMART Data into a Knowledge Graph for Learning Analytics

Abstract

The SMART team at the University of Liège aims to enhance learning ana- lytics by leveraging fine-grained student interaction data collected via the FB4You mobile application, which relies on the Experience API (xAPI). While xAPI provides a standardized format for recording learning activities, the resulting data is semi-structured, heterogeneous, and largely devoid of se- mantic coherence, creating significant challenges for analysis that require complex queries and extensive domain expertise. This thesis investigates the extent to which a knowledge graph can make implicit information in xAPI statements explicit, thereby supporting more effective learning analytics. We first analyzed over 2,100 heterogeneous xAPI JSON files to identify hidden relationships, undocumented conventions, and structural inconsistencies. To address these challenges, we developed PROV-EXT, an extension of the PROV-O ontology, and implemented RDF Mapping Language (RML) specifications to transform raw xAPI data into a semantically coherent knowledge graph comprising over 1.3 million triples. The utility of the knowledge graph was demonstrated through practical use cases, including student activity timelines, performance correlation analysis, and behavioral clustering. Results show that semantic modeling improves data interpretability, facilitates complex analytical queries, and enables prac- titioners to explore student behaviors and learning patterns more intuitively than raw JSON data allows. This work highlights the value of semantically enriched representations in learning analytics and provides a foundation for future enhancements, in- cluding ontology enrichment, scalability improvements, quality assurance, and user-friendly interfaces for educational practitioners.