Predicting future links between disjoint research areas using heterogeneous bibliographic information network

Abstract

Literature-based discovery aims to discover hidden connections between previously disconnected research areas. Heterogeneous bibliographic information network (HBIN) provides a latent, semi-structured, bibliographic information model to signal the potential connections between scientific papers. This paper introduces a novel literature-based discovery method that builds meta path features from HBIN network to predict co-citation links between previously disconnected literatures. We evaluated the performance of our method in predicting future co-citation links between fish oil and Raynaud’s syndrome papers. Our experimental results showed that HBIN meta path features could predict future co-citation links between these papers with high accuracy (0.851 F-Measure; 0.845 precision; 0.857 recall), outperforming the existing document similarity algorithms such as LDA, TF-IDF, and Bibliographic Coupling.