Graph Contrastive Learning Framework for Influential Node Ranking in Citation Networks

Identifying influential nodes in complex networks is a critical task with applications in social network analysis, information diffusion, and scientific impact assessment. The existing methods often rely on predefined centrality metrics, which are not generalize across different network structures. In this research work, we aim to develop a more scalable solution based on deep neural network. The proposed self-supervised framework based on Graph Contrastive Learning (GraphCL) combined with two-layer Graph Convolutional Network (GCN) to estimate node influence without requiring labeled data. The methodology initializes with citation networks, we apply graph augmentations through edge dropout and feature masking to generate diverse views of the input graph. A two-layer GCN is used then within GraphCL to learn node representations by optimizing a contrastive loss between augmented views. The obtained node embeddings capture the structural importance of nodes. The influence scores of nodes are computed as the norm of these learned embeddings, enabling unsupervised node ranking. We evaluate the proposed method on two benchmark citation datasets, Cora and CiteSeer, comparing it against centrality measures like Degree, PageRank, Closeness, Betweenness, and Eigenvector centrality. Experimental results show that the proposed approach achieves competitive NDCG score of 0.85 on the CiteSeer network and 0.77 on the Cora network, when compared with PageRank. When compared with closeness centrality, the NDCG scores of 0.83 on CiteSeer and 0.91 on Cora network. The proposed method results in a 48 % and 49% increase in NDCG scores on CiteSeer and Cora network respectively. The validation is performed using Susceptible-Infected-Recovery (SIR) model. These results underscore the scalablility, domain-agnostic nature, and robustness of the proposed method, offering a flexible foundation for influence estimation across diverse graph-based applications.