Identification of Influential Nodes With Topological Structure Via GRAPH Neural Network (GNN) Approach in Social Media Networks

The amount of data in social networks is vast these days, and constantly changing, making influential node identification crucial. The existing topologies are constantly changing due to the evolving behavior of the applied dataset. Node and leaf topology or feature-based value form the basis for machine learning and centrality computation. Hence, influential node value determination is based on the node attribute and network topologies. In the context of a large dataset, working towards the identification of the most influential node in the network, the Graph Convolutional Network (GCN) is the most effective and trusted approach. In the current research paper, the GCN has been projected as the most effective approach towards the identification of the node that is most influential in the graph-based dataset. The graph-based datasets are very large. A deep learning framework with the help of structural centrality via GCN, known as DeepInfNode, has been developed for the identification of the most influential Node. The Susceptible-Infected-Recovered (SIR) model is developed to identify the infection rate. This infection rate is further divided into three categories: Susceptible, Infected, and Recovered. The current approach uses the SIR model to collect contextual information to develop node representations. Higher values of F1 and AUC (Area under Curve) and F1 is visible when the suggested model is used. This has been discussed and explained in the experimental section. The observations prove that the above-mentioned strategy is precise and effective. It also suggests a potential new linkage within the network. An accuracy of up to 98% is achieved on all publicly available standard graphs available from Kaggle for different domains and datasets like Facebook, credit card fraud detection, Twitter, and Disease prediction using machine learning. Implementation of the proposed DeepInfNode works effectively and accurately for different domains. Additionally, performance improvement is confirmed during data processing and experimental analysis with the use of the DeepInfNode framework.