ABSTRACT
In informational recommenders, significant challenges arise from the need to handle the semantic and hierarchical structure between knowledge areas. This work aims to make advances toward building a semantically aware educational recommendation system, where the aim is to estimate the knowledge/interests of learners to leverage suitable recommendations. To do so, our proposed model incorporates notions of semantic relatedness between knowledge topics, propagating latent information across those semantically related topics. To introduce this novel learner model that exploits semantic relatedness, we make use of the Wikipedia link graph. Our final aim is to better predict learner engagement and latent knowledge in a lifelong learning scenario and evaluate how semantic knowledge graphs can facilitate this task. Our proposal, Semantic TrueLearn, one of the first attempts at fusing probabilistic graphical models and semantic knowledge graphs, is much more accurate than its non-semantic counterpart and builds a humanly intuitive knowledge representation, crucial in the context of education. Our experiments with a large dataset indicate that modeling semantic relatedness can improve user models that go beyond Semantic TrueLearn, showing the generalizability of our approach.
