Measuring Semantic Similarity of Learning Outcomes: Computational Approaches and Educational Implications

Learning outcomes (LOs) represent a pivotal component of contemporary instructional design, particularly within the context of modular and student-centred educational frameworks. However, the increasing adoption of flexible learning pathways, micro-credentials, and cross-institutional programmes has created a critical need for systematic methods to evaluate the semantic similarity between learning outcomes. This paper explores computational approaches to measuring such similarity, integrating symbolic, statistical, and hybrid methods. The SPIRAL model formalises LOs as structured, reusable entities aligned with Bloom's taxonomy. This study builds on the SPIRAL model by developing and testing a similarity metric that combines Sentence-BERT embeddings with an asymmetric adjustment based on cognitive level. A human evaluation study conducted in the domains of mathematics and physics provides a reference baseline for interpreting the results. The findings indicate that human raters consistently perceive higher similarity than the algorithm, especially in cases involving implicit conceptual or pedagogical relationships. This study makes three key contributions to the field: firstly, it advances the concept of semantic similarity as a pedagogical enabler of personalised learning pathways; secondly, it distinguishes between symmetric (equivalence) and asymmetric (implication) relations; thirdly, it promotes the use of similarity as a design driver in modular, interoperable course architectures. These contributions are instrumental in the development of intelligent educational systems that align with European frameworks, such as EQF and ESCO, while enabling scalable and meaningful personalisation of learning paths.