Node Degree Classification using Graph Modelling

Graphs serve as a potent tool for representing network data with nodes as objects and links as their relationships. The information within these graphs can be evaluated, retrieved, and visualized through Machine Learning (ML) techniques. However, using ML techniques with graphs is not so east because graphs are non-Euclidean structures and ML Methods are only suitable with Euclidean structures. To reduce this complication, the graph is encoded into a low dimension Euclidean vector representation, known as embedding vector, via the process known as node embedding. Recently, various Machine learning techniques had been developed that will automatically learns to encode the graphs in very fast and highly efficient way. Some of these techniques are, known as (GCNs). The aim of the project is to unfold: firstly, the objective of this study is to address the node degree classification and develop a model that can be utilized for node prediction on a high-precision dataset. to predict subject of a paper based on its words and citations network in an academic network that comprises citation information from top ML research papers. Initially, a homogeneous model is constructed utilizing GCN, GAT, and Graph SAGE techniques. To achieve node degree classification, a straightforward but efficient self-designed framework is introduced by implementing and evaluating three different architectures (GCN, GAT, and Graph SAGE) in PyTorch Geometric. For node prediction, the most precise framework with GNN is employed, and ensemble learning is utilized to improve performance. Detailed experiments conducted to establish robustness and effectiveness of the proposed solution