CAN-SIN: A Cross-Layer Heterogeneous Academic Network with Semantic Information

In this paper, we focus on incorporating the semantic information into the structure of academic networks to enrich the dimensionality of extracted features. We propose a cross layer scholar-paper network that can capture the characteristics of heterogeneous academic networks. In addition, we leverage the BERT model, which is widely used in the realm of natural language processing (NLP), to integrate the semantic information of the scholar papers. We also introduce a new concept, “close collaborator to tackle data leakage issues. This can be used in many downstream tasks such as automatic detection of conflict of interests among scholars. Extensive experiments on two datasets show that our enhanced cross-layer model is both effective and lightweight, and outperforms three strong baselines. Further analysis shows that our model successfully combines the semantic information and the topology of the whole network.