Influential Node-Based Random Walk Graph Embedding with Graph Neural Networks in Complex Graphs

Centrality Metrics are widely used metrics for graph analysis, which involves describing the significant nodes in terms of connected nodes, mostly positioned in all geodesic paths and often connected to higher degree nodes. It is crucial to various applications, including Personalized Recommendation, Community Detection, and Network Split. Computing these metrics on scale-free graphs is a challenging factor about time complexity. To compute these social metrics, this research study considers a generalized form of all neural models called Graph Neural Network (GNN). A GNN is a class of deep neural models for graph processing. GNNs perceive two functions: aggregation and combination. Aggregator controls the internal structure through neighborhood aggregation by considering each node's embedding feature later, the combination function joins this with the node itself. The proposed work uses different embedding methods like Node2Vec, DeepWalk, Stuct2Vec, and Random walk with a restart to map graph structure into vector space for prediction. As random walks are a demanding task for these algorithms, in this work a proven computationally less approach is used. By training on well-known data, the proposed model produced random walks proportionally with the structural features. Experiments on random scale-free graphs prove that the proposed model attains accuracy without scarifying the computational speed.