Link List Predection using Graph Neural Network

Abstract

In this study, the applications of Graph Neural Networks (GNNs) for link prediction in citation networks have been explored using a dataset called Cora. We can visualize the networks as paper nodes of the graph and the citations as edges of the graph, and we can form a complex graph that can be analyzed using the GNN model. Previously, we used the Stallergraph library to perform some link prediction tasks using some necessary libraries to construct, validate, and train. This is based on the accuracy and the training loss. We achieve a satisfactory result of 0.7771 in predicting the existence of links between nodes. We understand the complex relationship of the graph, and the losses show a consistent and stabilized decline in losses. The loss is 0.9994. We can validate the model by learning the patterns of this complex network. Also, we discuss the challenges, but mainly, we have studied the step processes of how the graph convolution works. In deep learning for having a huge data, it is a must to experience a latency-sensitive structure where previously we used arrey to solve problems of linking and sorting. Link lists present models of the the next generation for handling complex data. Within a graph adgeny we can find different features that previously could not be fetched or not used to validate, but with the help of neural convolution, we can visualize a real-time connection and link within different classes.