Ontology-based semantic relatedness for detecting the relevance of learning resources

Abstract

Semantic relatedness measures have proven useful for a number of applications, such as querying personalized web resources, word sense disambiguation, or real-word spelling error correction. Most semantic relatedness measures between concepts are based on the concept hierarchy of a domain ontology. In this article, we propose a semantic relevance (SR) measure that expresses the semantic relatedness between a learning resource and the learning context of a learner. In our case, both the learning resource and the learning context are described by graphs using the learning concepts of the domain model. Our SR measure aims to detect the relevance of a learning resource for a particular learning context of a learner. In our work, the SR measure is based on the assignment of relative weights to the learning concepts describing the learning resource according to their relationships with the current concept of interest to the learner. The proposed measure achieves better results than the relatedness measures from the literature and yet is much simpler than most of them. It is shown to achieve a correlation of from 0.627 to 0.945 with expert ratings. This measure is implemented and used in a learning organizer, a system which generates adaptive hypermedia courses and reuses learning resources from distant web repositories, called Organisateur de Parcours Adaptatifs de Formation (OrPAF). In OrPAF, learning resources are annotated in order to be queried for a particular learning context, which is represented by a map of annotated learning concepts, called an adaptive conceptual map. The proposed SR measure is used in order to automatically detect the learning resource relevance.