Abstract
The article considers the problem of constructing personalized educational recommendations based on a heterogeneous knowledge graph that combines data from university curricula, vacancies, and online courses. To rank courses by relevance to the user’s competence, an approach based on heterogenous graph neural networks is proposed. The experiments were conducted on a heterogeneous graph that includes different types of nodes and edges. In the experimental study, the quality of the models was assessed using different metrics. The results showed the superiority of the recommendation model with the HGTConv layer on the link task weight prediction, especially in terms of ranking quality metrics, which indicates its suitability for building recommendations in educational systems. The presented approach can be used to develop adaptive recommendations in education focused on the user’s career goals.