Knowledge aware intent learning and contrastive learning for course recommendation
Tianci Zhu, Yingying Yao · 2024
In recent years, online education has developed rapidly, providing a vast array of educational resources. Course recommendation algorithms are crucial for MOOC platforms, as they can alleviate the issue of information overload to a certain extent. Due to the cold start problem in recommendation algorithms, researchers have introduced Knowledge Graphs (KG) to address this issue. However, many course recommendation algorithms based on Knowledge Graphs overlook the learning intentions of students when selecting courses, which leads to a decrease in recommendation accuracy. Additionally, the introduction of information such as Knowledge Graphs can bring noise, reducing the performance of the recommendation system. In response to these challenges, We proposes a course recommendation algorithm called Knowledge Graph Intent Network and Multi-graph Contrastive Learning (KGIMCL).This paper delves into the modeling of the interaction intentions between students and courses based on the Knowledge Graph Intent Network, employing attention mechanism strategies to more effectively measure the importance of different relationships within the Knowledge Graph. Furthermore, this paper introduces a novel multi-graph contrastive learning approach. Specifically, it contrasts the course embeddings generated by the Knowledge Graph Intent Network with those generated by the course semantic relationship graph. This contrastive learning strategy enhances the model's performance by reducing the noise introduced by additional knowledge information while incorporating more knowledge into the recommendation system. This paper provides a perspective for MOOC platforms when developing algorithms, considering the learning intentions of students when selecting courses, and reducing the noise that additional information such as Knowledge Graphs might cause in the recommendation system through contrastive learning. The effectiveness of this model is validated on a public MOOC dataset.