← Back to Publications List

Graph-Based Approaches to Social Network Analysis: Pinpointing Central Nodes

Students & Supervisors

Student Authors
Ahammed Ekrak Hossain Utsha
Bachelor of Science in Computer Science & Engineering, FST
Akash Majumder
Bachelor of Science in Computer Science & Engineering, FST
Shadman Sakib Faruki
Bachelor of Science in Computer Science & Engineering, FST
Supervisors
Sharafuddin Mahmood
Assistant Professor, Special Assistant [osa], FST

Abstract

The process of recognizing powerful nodes in intricate social networks is critical in knowing the flow of information and stability of the system. In the paper, a hybrid framework is suggested to combine both the conventional graph-theoretic metrics and machine learning and current state-of-the-art Graph Neural Networks (GNNs). Based on a Barabasic-Albert scale-free network, 14 topological characteristics, such as degree, betweenness and eigenvector centrality, were mined to measure the importance of nodes. We use a multi-model fusion model that compares classical Algorithms such as the Use of Before and After Random Forest and gradients boosting with the more sophisticated GNN architecture such as the use of GraphSAGE and Graph Attention Networks (GAT). The experimental findings prove that although the Random Forest model had the highest F1-score of 0.9565, the graph sage model had better structural generalization with an AUC of 0.9992. Moreover, the GNN Ensemble (GAT + GraphSAGE) was strong as it improved the predictive stability and deeper structural dependence. The fact that the top 20 influential nodes that were identified have an average confidence score of over 0.85 ascertains the reliability of the framework. All in all, this hybrid model offers a scaleable, interpretable, and highly accurate way of identifying the influencers within dynamic social setups.

Keywords

Social Network Analysis Influential node detection Graph theory Graph neural network (GNN) Machine learning Centrality metrics Barabasialbert model Graph sage Graph attention network (GAT) Random forest Gradient boosting Ensemble learning Network visualization Information diffusion Data analytics.

Publication Details

  • DOI: https://doi.org/10.1109/QPAIN69676.2026.11546297
  • Type of Publication:
  • Conference Name: IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence, and Networking (QPAIN)
  • Date of Conference: 16/04/2026 - 16/04/2026
  • Venue: Chittagong University of Engineering and Technology (CUET)
  • Organizer: IEEE Photonics Society Bangladesh Chapter
View on Publisher