A Personalized Course Recommendation Model Integrating Multi-granularity Sessions and Multi-type Interests

The open online course (MOOC) platform has seen an increase in usage, and there are a growing number of courses accessible for people to select. An effective method is urgently needed to recommend personalized courses for users. Although the existing course recommendation models consider that users' interests change over time, they often model users' learning records as a single time-granularity sequence and ignore the collaboration between different time-granularity sessions when recommending courses. In addition, most course recommendation models tend to use the deep network, which weakens the memory ability of the model. Few methods simultaneously consider long and short-term interests and individual course interests in the latest session, which results in a decline in model performance. To resolve these problems, we design an innovative personalized course recommendation model that Integrating Multi-granularity Sessions and Multi-type Interests (IMSMI), which converts user-course interaction sequences as multi-granularity sessions and uses different types of attention mechanisms to capture multi-type interests. Meanwhile, we introduce the residual connections to further strengthen the memory capability of IMSMI. Experimental results using the XuetangX dataset available to the public demonstrate that IMSMI significantly surpasses other competing models on evaluation metrics. Compared to the next best model, Recall@3 is increased by 20.50%, and MRR@3 is increased by 18.07%.