Dynamic Feature Collaborative Variational Auto-Encoders for Academic Paper Recommendation

The rapid growth of scientific research has contributed to an overwhelming surge of information in academic papers, and scholarly paper recommendation systems have developed rapidly. However, existing systems often underutilize valuable feature information and fail to consider changes in the attractiveness of papers over time. Therefore, this paper proposes a novel approach called the dynamic feature-based collaborative variational auto-encoders (DFC-VGAE) model for the academic paper recommendation. In this study, we preprocess the features using a pre-trained natural language model and capture structural features through co-authorship and citation networks. The model also modifies the encoding and decoding. Promising results were achieved through a real-world experiment using a comprehensive academic paper recommendation dataset.