Exploring the Efficiency of Various GNN Architectures for Node Classification in Social Networks

Node classification is a crucial task in many applications, where the objective is to accurately classify nodes based on their features. Unlike traditional machine learning methods, which often fall short in handling graph-structured data, Graph Neural Networks (GNNs) offer a robust solution to this problem. This research aims to identify the most effective GNN model for node classification by evaluating several methods, including Graph Convolutional Networks (GCN), GraphSAGE, Graph Attention Networks (GAT), and ChebNet. Through careful tuning of key hyperparameters, specifically the number of hidden units per layer and the learning rate, the study achieved optimal model performance. The best results were obtained using the GraphSAGE model with 64 hidden units per layer and a learning rate of 0.01, leading to exceptional accuracy and F1 scores of 0.999823 and 0.999852, respectively. These findings demonstrate the effectiveness of hyperparameter optimization in enhancing GNN performance for node classification tasks.