Link prediction based on directed network clustering coefficient

In this paper we analyse network link prediction methods based on a concept of digraph clustering coefficient proposed by M. Bloznelis and L. Leskelä. The goal of this research is to define new link prediction indices derived from the clustering coefficient mentioned above. We also aim to empirically evaluate the performance of the new indices when solving link prediction problems within real-world networks. In this paper we first define the link prediction problem and analyse possible methods of evaluating its results. We also explore and compare several already widely known link prediction methods. In the second and the third chapters we thoroughly investigate the already mentioned digraph clustering coefficient and define three new link prediction methods: the witness, the common-targets, and the indegree-aggregation indices. In the fourth chapter of this paper we describe data and networks built for our experiments. Lastly, we carry out the experiments with an aim to empirically evaluate link prediction quality of the newly defined indices. Based on the experiments we have concluded that the witness and the common-targets indices result in high quality link prediction. In the meantime the indegree-aggregation index has shown different results - network links are predicted the opposite way. That means the links the index suggests as the existing ones are those that actually do not exist and vice versa