融合多种时间关系的时序图课程推荐算法OA

In the learning process,the temporal features of learning records reflect a variety of important information,such as learners' chang-ing interests,learning cycles,and successively dependent relationships between courses.At present,the course recommendation only consid-ers the course order relationship,and most of the graph neural network course recommendation algorithms completely discard the temporal fea-tures,resulting in performance degradation.A time-series graph integrating multiple temporal patterns for course recommendation is proposed to make full use of temporal features to improve the representation accuracy.In order to obtain fine-grained time information,the model first converts the temporal features into three kinds of time patterns:absolute time,sequential time and interval time.Secondly,the model con-structs a learner-course interaction time-series graph,assigns individualized aggregate weights to neighbor nodes through three kinds of time patterns embedding and attention mechanisms,and then obtains learner and course representations through residual connection and multi-lay-er propagation for final prediction.A large number of experiments on the MOOCCourse dataset show that the proposed model outperforms other advanced recommendation models by 6.58％and 2.61％on R@5 and NDCG@15,respectively,and prove that the performance of combining three time patterns is better than considering only the sequential relationship of courses on R@5 and NDCG@15.