In recent years, the importance of diversity has increasingly been acknowledged in many fields. In this context, the diversity of links between nodes in social networks is also important. In this study, I propose a method to maintain the diversity in network when making link predictions. To this end, drawing on prior studies, I define a new index of the diversity as well as an index based on similarity. These indices are then introduced into a graph neural network to predict network links. The results of an experimental application of this method on real data confirm that it maintained the diversity while exhibiting high prediction accuracy. I expect this approach to be applied to improve the performance of recommendation systems.
