Title: Graph classification system using normalised graph convolutional networks

Authors: G. Naga Chandrika; E. Srinivasa Reddy

Addresses: Department of Computer Science and Engineering, ANU College of Engineering and Technology, Guntur – 522510, AP, India ' Department of Computer Science and Engineering, ANU College of Engineering and Technology, Guntur – 522510, AP, India

Abstract: Recent years have authenticated a dramatic increase in graph applications due to their advancements in their informative and social connectivity. In large-scale networks, graph data contains huge information and exhibits distinct characteristics. Traditional graph convolutional networks cannot handle the problem of covariate-shift in the neural-networks during analysing the patterns. This research presents an intelligence-based graph-classification model for citation networks by using normalised graph convolution networks to handle covariate-shift problem and provides greater regularisation with a decrement in loss to a greater extent. This problem is vanquished by utilising the normalisation constraint for the individual batch constructed as per network developed for graph-structures which eventually lead to building a robust model. The model classifies the underlying connectivity patterns of the structural and informative relationships with appropriate embeddings. This information is fed to neural-network and learns hidden layer representation that encode both local graphs and features of the nodes in the citation network.

Keywords: citation network system; classification; document classification; entity classification; graph convolutional networks; covariate shift.

DOI: 10.1504/IJSSE.2021.121461

International Journal of System of Systems Engineering, 2021 Vol.11 No.3/4, pp.320 - 335

Received: 08 Jul 2020
Accepted: 03 Oct 2020
Published online: 14 Mar 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article