Concept Enhanced Content Representation for Linking Educational Resources

The education sector has been undergoing a welcoming change in recent years with the introduction of a wide variety of digital content openly available to students. Due to the volume of this new digital content, it is very difficult for learners to find the needed information at the right time. Digital textbooks, as well-curated domain knowledge sources, could provide a conceptual and physical platform that unites disparate educational resources as one entity. Educational resource linkage, with state-of-the-art techniques, are based on term-level and topic-level representations. However, term-level representations suffer from the term-mismatch problem and often, topics are too broad for linking to other educational resources. To address these challenges, we propose to link educational resources through concept-level representation. The proposed model generates concept embeddings by utilizing domain-specific educational content and external knowledge graph resources to achieve robust and effective concept-level representations. We conducted evaluations of the proposed models on multiple contents linking tasks, and the results demonstrate that concept-level representations perform better than the state-of-the-art representations in helping students to find more learning resources easily. This could increase both students' learning and satisfaction.