Analysis of Nodes and Edges Types in Large-Scale Networks

Abstract

In networks, node and edges can occupy different roles and types. There can be nodes with significant influence on other nodes or the whole network than other nodes, but it not always trivial task to identify them in large scale networks. The same applies for the connections between nodes. With analysis, we can better understand the network dynamics and impact which can these nodes or edges potentially have. This thesis is divided into two main sections.

In the first section, the analysis of node and edge types is performed using graph neural networks, which is a relatively new architecture design to handle network data. We perform the classification of node roles and edge-ties types using graph neural networks. They are a relatively new approach to performing machine learning on networks and have great potential to generalize the relations between nodes and edges in large scale networks.

In the second section, we introduce a novel network embedding method based on nonsymetric relationships. Its aim is to embed all necessary information in vectors, on which analysis with other tools can take place. We also conduct experiments trying to measure the quality of the embedding, which is not a trivial task. Even if the results are not the best compared to other embedding algorithms, the results are still very comparable, and it can be used as efficient tool for analysis of community structures.