Student Course Recommendation System Using Semantic Aware Temporal Gating With Time Transformer

Abstract

Course recommendation system works by analyzing users’ interests and recognizes their historical behavior such as course enrollment and previously watched videos to recommend relevant courses. However, the existing course recommendation systems in Massive Open Online Courses (MOOCs) face challenges in capturing complex semantic information, which impacts student requirements with an unsatisfied recommendation. By combining historical behaviors without temporal awareness, the model does not provide a proper recommendation and it lacks sequential dependencies. This research proposes Semantic Aware Temporal Gating (SATG) with Time Transformer (TT), which dynamically weights historical learning interactions by cross-referencing elapsed time with content-based Knowledge Graph (KG) semantics. The SATG helps capture sequential dependencies between historical user behaviors; hence, the relevant course is recommended by the model. Additionally, SATG preserves long-term, semantically relevant connections while suppressing irrelevant past behaviors. The proposed SATG-TT achieves a Normalized Discounted Cumulative Gain (NDCG) of 54.82 and 51.02% at @5 as well as @10 on the MOOCCube and XuetangX datasets, respectively. These results demonstrate that the proposed model exhibits competitive performance compared to existing baseline models such as Knowledge Sequence Course Recommendation (KSCR).