Attribute-Aware Graph Recurrent Networks for Scholarly Friend Recommendation Based on Internet of Scholars in Scholarly Big Data

Abstract

The academic society is stepping into the age of scholarly big data, where finding suitable scholars for collaboration has become ever difficult. Scholarly recommendation approaches are designed to overcome the information overload problems. However, previous methods mainly consider network topology without considering scholars’ academic information and the manually designed similarity measurements may not have a good performance when applying to large-scale sparse networks. To this end, this paper proposes to design a scholarly friend recommendation system by taking advantages of network embedding and scholar attributes. It is worth mentioning that different from traditional scientific collaborator recommendations, our goal is to recommend potential friends for scholars using academic social networks. We first construct an attributed social network by extracting scholars’ academic attributes from digital libraries. Then, we perform an attributed random walk which can jointly model network structure and scholar attributes. Finally, a novel graph recurrent neural framework is adopted to embed attributed scholar interactions within the model for recommendations. Experimental results on two real-world scholarly datasets demonstrate the effectiveness of our proposed method