Knowledge-Aware Attentive Embedding Learning for Educational Group Recommendation

Abstract

Abstract With the advancement of personalized education, the use of recommendation to improve the effectiveness of cooperative group learning has gradually become one of the core methods in education. However, most existing group recommendation methods focus on aggregating the interests of members within a group, ignore the collaborative relationships between groups, members, and items. To address the above problem, we propose an approach called Knowledge-aware Attentive Embeddings learning for educational group recommendation(KAEG), in which we construct a knowledge graph to capture various information and collaborative relationships for group decision modeling. Specifically, the method designs Simplified Graph Convolutional (SGC) networks in the knowledge graph to capture the abundant structural information of items and the similarity of interests between users and between groups. Based on this information and similarities, we design attention-embedded aggregators to learn information about collaboration between members and groups for modeling group preferences. In addition, we use a joint training strategy to train the user recommendation task and the group recommendation task in the same process, which makes them mutually reinforcing. Experiments on real-world datasets demonstrate that our KAEG model outperforms other state-of-the-art baseline models.