Tsourekas Stergios (Phd Candidate)

Thesis title: Analysis and Optimization of Centrality Metrics in Reference Graphs and Social Networks

Supervisor: Sidiropoulos Antonis

Advisory Committee Members:

Antoniou Efstathios, Professor, Dept. of Information and Electronic Engineering, IHU
Ougiaroglou Stefanos, Assistant Professor, Dept. of Information and Electronic Engineering, IHU

Abstract:

This doctoral thesis focuses on the in-depth study, evaluation and application of centrality metrics in citation graphs and social networks, with the aim of understanding the importance of nodes in complex and dynamic information structures. Despite the widespread use of metrics such as Degree, Betweenness, Closeness, PageRank and Eigenvector centrality, there is no unified framework that guides the selection of the appropriate metric depending on the nature of the network and the intended goal. The complexity and time evolution of real networks, such as citation graphs or social networks, necessitate the development of innovative methods that combine graph theory with machine learning techniques.

The current state of the art review identifies advances in techniques such as Graph Neural Networks (GNNs), which are used to learn node representations and efficiently estimate centrality even in environments with incomplete or noisy data. The dynamic and multi-level structures, as well as the heterogeneous characteristics of modern graphs, require the adaptation of traditional metrics or the development of new ones, such as temporal and higher-order metrics. At the same time, the need for methods that are scalable, interpretable, and flexible, capable of operating on a variety of data types, is highlighted.

The proposed research approach includes the analysis of metrics in different graph categories, the creation of a theoretical classification framework, the development and evaluation of new hybrid models, and the creation of a “centrality recommender” tool. Particular emphasis is placed on the use of meta-learning and reinforcement learning for the automatic selection of the appropriate metric, as well as the construction of open source software to enhance repeatability and the transfer of know-how to the scientific community.

The thesis seeks to contribute to the theoretical deepening and practical application of the concept of centrality in real large-scale data, enhancing the ability to understand, predict and manage influence and information in network environments.