Bilateral knowledge graph enhanced online course recommendation

Recommender system can provide users with items that meet their potential needs in mass information. Its development provides new ideas and supporting technologies for applications in online education scenarios. The previous recommendation methods usually only consider the enhancement of the item side, but ignore the importance of the user characteristics to the recommendation, and are not suitable for the online education scenario. To address this problem, we take knowledge graph as the auxiliary information source of collaborative filtering and propose an end-to-end framework using knowledge graph to enrich the semantics of the item representation. In particular, faced with the thorny problem of cold start, the framework makes use of the static features of users to personalize the modeling of new users. Experimenting with two public datasets and an industrial dataset, we demonstrate that the framework has significant performance improvements over the baseline and can maintain satisfactory performance with sparse user–item interactions. •A personalized KG-enhanced model is proposed for online course recommendation.•An attribute-level attention network is used in course modeling.•The user graph is constructed to find users with similar static features.•The interactions of similar users are used to simulate the preference of a new user.•The model can effectively alleviate data sparsity and cold start problems.