Publication recommendation in incomplete networks based on graph learning

Tremendous academic information in scholarly big data brings problems of information overload and information loss. Scientific paper recommendation systems are developed to help with the problem of information overload by recommending relevant papers to researchers. However, few recommender systems focus on the problem of missing information. To tackle this problem, we develop a scientific paper recommendation system, namely INSURE, to recommend papers in an incomplete network. INSURE can mine the implicit relationships between nodes, and infer the original network structure and attributes from the observed incomplete network. Hence, the problem of confusion results caused by missing structure and attributes can be solved. At the same time, INSURE takes into account both structural similarity and textual similarity of the paper nodes in the recommendation process. In addition, INSURE also considers factors affecting paper quality during random walks to recommend high-quality studies when citations have not accumulated. Experimental results on the CORD-19 dataset show that INSURE outperforms baselines in terms of recommendation performance and recommendation quality.