STR-SA: Session-based Thread Recommendation for Online Course Forum with Self-Attention

In recent years, along with the rapid development of Massive Open Online Courses (MOOCs), a large number of students participate in MOOC courses. MOOC course forum is very important for students discussing questions related to online courses when they don’t have enough offline resources to get help from. There are a lot of qualified discussion threads in MOOC courses which are helpful for understanding the course materials. However, students are always overwhelmed by the enormous amount of threads. A solution to alleviate this problem is to recommend a list of threads for each student that she/he may be interested in. Existing thread recommender systems are insufficient to capture the complex relationships among the threads in a student’s visit session, and do not simultaneously take into account the student’s global preference, as well as her/his current interest in that session. In this work, we propose a novel neural network framework for session-based thread recommendation (STR-SA for short). The proposed method, which recommends threads to a student based on the threads viewed by the student in the current session, learns the relationships among threads by applying self-attention mechanism. Furthermore, we capture the global preference of the student by combining the viewed threads, and consider the latest-viewed thread as the current interest of the student. Extensive experiments conducted on three course-forum datasets show that the proposed model STR-SA significantly outperforms other representative methods for MOOC thread recommendation.