Graf Vector Representation Methods for Machine Learning Problems

Abstract: Graphs, such as social networks, communication networks, are nat- urally used in various real-world cases. By analysing them one can obtain an un- derstanding of the structure of society, of a language and different communication models. Many approaches were suggested for the analysis. Recently, methods that use the representation of graph nodes in the vector space have become widespread in the scientific research community. This paper presents a structured analysis of the various methods of attachment (embedding) of graphs proposed in the litera- ture. First of all, three methods of graph embedding (LLE, Laplacian Eigenmaps, Node2vec) are considered, after which their use is demonstrated by the example of the «Karate club» graph. Next, we will consider the dependence of the accuracy of clustering depending on the probability of connection between clusters.