PretoTMLP: A Novel Approach for Link Prediction on Bibliographic Networks Using Pretopology and Text Mining

In this paper, we propose a novel approach for link prediction on bibliographic networks using pretopology theory and text mining. Supervised learning method is used for link prediction, in which the highlights are to extract the new features from analyzing the network’s structure using pretopology theory and discovering textual content by combining text mining methods. The pretopological features are built by using pseudo-closure distance based on the pseudo-closure function in binary relation space and valued relation space. This approach can capture the complicated neighborhood set of a set in complex networks that cannot be solved by graph theory. Additionally, we analyze content features on a larger text corpus compared with preceding research incorporating text mining methods. Finally, machine learning and deep learning are employed with the consolidation of pretopological features and content features to identify link formation. Our proposed model was conducted on real-world datasets of the bibliographic network and demonstrated superior efficiency compared to existing methods.