Similarity Metrics and Visualization of Scholars Based on Variational Graph Normalized Auto-Encoders

With the rapid development of Internet technology and academic social media, it has become one of the important research directions of social networks to measure the similarity between scholars, which enables users to find scholars with similar research interests or similar backgrounds in the scholars’ network and promote scientific research cooperation and communication. In this paper, we propose an algorithm for measuring the similarity of scholars based on variational graph normalized auto-encoders (VGNAE), which fuses the social relationship and attribute information of scholars, obtains the implicit feature representation of scholars through graph convolutional neural network encoding, and then compares different implicit feature vectors to measure the similarity of different scholars. In order to evaluate the effectiveness of this method, the SCHOLAT social network dataset is selected for experiments in this paper. The experimental results show that the model achieves the best performance on both the AUC and AP metrics on the task of measuring scholar similarity, with 98.7% and 98.8%, respectively, relative to other traditional widely used algorithms. In addition, by fusing the scholar similarity information into the scholar network visualization, users can access the connection between different scholars more intuitively and efficiently.