Knowledge Concept Recommender Based on Structure Enhanced Interaction Graph Neural Network

Online education is becoming more and more popular. Although there are many online courses, this brings users trouble: (1) courses with the same name have different emphases; (2) there is dependence between courses in learning order. These phenomena make learners spend a certain amount of energy to find suitable courses, reducing their interest in online education. To help learners better conduct online learning, we study the problem of knowledge concept recommendation. This paper proposes a knowledge concept recommendation model based on a structure-enhanced interactive graph neural network (KCRec-SEIGNN). As for user representation, multiple entities in the knowledge concept recommendation scenario are organized into a heterogeneous graph and then graph convolution based on meta-path guidance to learn user entity representation on the heterogeneous graph. As for knowledge concept representation learning, we capture the co-occurrence information of knowledge points in the interaction sequence by using the knowledge concept interaction sub-sequence as a wizard and then build a knowledge concept interaction graph using the co-occurrence information. When aggregating the neighbor node information, we retain the structure information between the neighbor node and the target node and use an attention mechanism to distinguish the contributions of different neighbor nodes. Lastly, input the representation of user entities and knowledge concepts into a rating layer based on extended matrix decomposition to recommend knowledge concepts by rating. We conduct a series of Experiments on public real-world datasets, XuetangX, showing that the model is more effective for knowledge concept recommendation than some of the latest methods.