Collaboration prediction in heterogeneous academic network with dynamic structure and topic

Academic collaborations improve research efficiency and spur scientific innovation. However, scholarly big data has hindered scholars from finding suitable collaborators. Although some studies have involved the prediction problem of academic collaborations, they neglect the rich dynamic information of the heterogeneous academic network. In this paper, we propose a prediction model for academic collaborations, which considers both the dynamic structure and content information. We first formally define the dynamic academic network and the collaboration prediction problem. Then, a scholar representation model is designed by capturing both the dynamic structure and content features, together with the macro-impact of overall academic trends. Finally, we build the prediction model based on the representation result of scholars. Extensive experiments for predicting new collaborations are conducted on the DBLP dataset. The experimental results on the accuracy, F1, and AUC metrics demonstrate that our method outperforms the baseline methods and can predict academic collaborations efficiently.