In scientific social networks, academic groups play a significant role in facilitating collaboration between researchers and promoting the dissemination of papers, which provides unique opportunities for paper recommendation. However, existing paper recommendation methods rarely consider the valuable group information, which limits their potential for improving recommendation performance. In this article, a novel multi-graph fusion network with attention mechanism (GI-MFA) is proposed for paper recommendation considering group information. First, the group-researcher bipartite graph, the researcher-paper bipartite graph and the group-paper bipartite graph are constructed to model the relationships between researchers, papers and groups. Graph neural networks are used to learn the embeddings of researchers and papers at both the individual and group levels across these bipartite graphs. Second, to effectively fuse the individual-level and group-level embeddings, we introduce researcher-wise attention and paper-wise attention mechanisms. To verify the effectiveness of GI-MFA, experiments are conducted on a real-world dataset CiteULike. The experimental results demonstrate the superiority of GI-MFA over all baselines.
Wang et al. (Tue,) studied this question.