A hierarchical RNN-based model for learning recommendation with session intent detection

Since the emergence of MOOCs (Massive Online Open Courses) in the last decade, online education continuously evolves. With the abundance of learning resources provided by MOOC platforms, recommender system can be used to personalize learners' learning experience with respect to learning material consumption. To provide user-adaptive and beneficial recommendation result, the recommender system should be designed with respect to properties of the online learning context, especially the sequential property of learning behaviors. In this paper, we propose a novel model SOLR, a session-based sequential model for online learning material recommendation. We use hierarchical RNN to model online learners' learning sequences on both in-session and cross-session levels. Additionally, attention mechanism is used within sessions to model users' learning session intent. The model is able to learn a hierarchical representation of users' long-term learning history as well as short-term session sequential patterns. We conducted comparative experiments with session-based recommendation baseline methods as well as an ablation study on real-life MOOC dataset. The results show that our model achieves better recommendation results and provide justification for the sequential modeling and model training mechanism implemented in our model.