Full Learning Path Recommendation Based on Temporal and Co-Occurrence Relations

Full learning path recommendation aims to recommend a fixed-length sequence of concepts to a learner, thereby enhancing his or her understanding of target concepts. While existing research has successfully recommended learning paths based on temporal relations, it struggles with co-occurrence relationships due to two main challenges: (1) How to balance the roles of temporal and co-occurrence relationships. (2) How to model temporal and co-occurrence relationships. To tackle these challenges, we propose a Learning Path Recommendation Framework Based on Co-occurrence and Temporal relations (RFCAT). Specifically, the framework first encodes concept graphs and learning resource feature matrices in an encoder. Secondly, the temporal graph convolutional neural network (T-GCN) of a decoder module realizes the knowledge level prediction task. Finally, the KT module generates the optimal full learning path. To evaluate the performance of our method, we extract co-occurrence relations on 4 datasets and conduct sufficient experiments. The results demonstrate the superiority and effectiveness of RFCAT.