Link prediction in co-authorship networks based on hybrid content similarity metric

Link prediction in online social networks is used to determine new interactions among its members which are likely to occur in the future. Link prediction in the co-authorship network has been regarded as one of the main targets in link prediction researches so far. Researchers have focused on analyzing and proposing solutions to give efficient recommendation for authors who can work together in a science project. In order to give precise prediction of links between two ubiquitous authors in a co-authorship network, it is preferable to design a similarity metric between them and then utilizing it to determine the most possible co-author(s). However, the relevant researches did not regard the integration of paper’s content in the metric itself. This is important when considering the collaboration between scientists since it is possible that authors having same research interests are more likely to have a joint paper than those in different researches. In this paper, we propose a new metric for link prediction in the co-authorship network based on the content similarity named as LDAcosin. Mathematical notions of the link prediction in the co-authorship network and a link prediction algorithm based on topic modeling are proposed. The new metric is experimentally validated on the public bibliographic collection.