Combining educational resources using graph representation learning

Open Educational Resources (OERs) are teaching, learning, and research materials destined for the public, allowing them to be freely used. They can be used by teachers to create new courses. Teachers can combine different OERs to achieve a specific learning objective. The CLARA project was launched to empower teachers by facilitating the creation of licensable educational resources based on existing ones.
In this thesis, funded by the CLARA project, our goal is to enrich the CLARA educational corpus of OERs with useful relations between them thus facilitating the navigation for the teachers. In order to do so, several contributions were made in this thesis. First, the creation of a dataset construction tool that allows users to create their own tailored educational datasets from YouTube video transcripts. Second, the development of an OER-specific vector representation (embedding) that considers the specificities of OERs which are content-centrality and the presence of semantic features. Third, the proposal of a querying method that retrieves OERs relevant to a list of keywords based on OER representations. Fourth, the design of a model that identifies possible precedence relations between pairs of OERs by using Knowledge Graph (KG) and exploiting a Graph Neural Network (GNN).
The contributions made work in harmony in order to enrich the CLARA corpus with educational resources, retrieve them and identify possible relations between them facilitating the navigation for the users.