2025 Asian Conference on Communication and Networks (ASIANComNet 2025)

The web-scale social networks, e-commerce platforms, knowledge graphs, and search engines have a glaring need for advanced processing techniques. Unlike traditional machine learning and deep learning models, the web-scale social networks and e-commerce platforms have complex structures that are best understood as graphs. In this paper, we study the use of Graph Convolutional Neural Networks (GCNNs) to improve the processing of web-scale data. With the aid of GCNNs, data that is graph-based can be schemed based on it unique connections and relation, allowing for enhanced performance in node classification, link prediction, and content recommendation. We propose a GCNN that can perform optimally while managing the size and sparsity of true-world web graphs. Incorporative of sampling methods, mini-batch training, and parallel processing, our architecture can be computed efficiently in distributed systems. GCNNs are proven to improve node and edge representation, for enhanced performance in tasks of node classification, link prediction, and content recommendation. In terms of accuracy, scalability, and generalization, citation networks, user-item graphs, and web link structures demonstrates that our technique surpasses greater than traditional models. Besides, we look into additional aspects of the model’s interpretability, considering memory optimization and inference in real-time. Our research underscores the profound impact GCNNs have on the development of intelligent and scalable solutions for future web-scale systems.
 

Graph Convolutional Neural Networks, Web-Scale Data, Scalable Graph Processing, Node Representation Learning