Representation Learning of Knowledge Graph for Wireless Communication Networks

TL;DR

This paper constructs a knowledge graph from wireless communication data and employs a graph convolutional neural network to classify nodes and predict relations, enhancing wireless network analysis and anomaly detection.

Contribution

It introduces a novel graph-based representation learning model tailored for wireless communication data, improving node classification and relation prediction accuracy.

Findings

The model outperforms existing unsupervised graph neural networks like VGAE and ARVGE.

It achieves automatic node classification and anomaly cause tracing.

The approach is validated on public datasets with superior results.

Abstract

With the application of the fifth-generation wireless communication technologies, more smart terminals are being used and generating huge amounts of data, which has prompted extensive research on how to handle and utilize these wireless data. Researchers currently focus on the research on the upper-layer application data or studying the intelligent transmission methods concerning a specific problem based on a large amount of data generated by the Monte Carlo simulations. This article aims to understand the endogenous relationship of wireless data by constructing a knowledge graph according to the wireless communication protocols, and domain expert knowledge and further investigating the wireless endogenous intelligence. We firstly construct a knowledge graph of the endogenous factors of wireless core network data collected via a 5G/B5G testing network. Then, a novel model based on graph convolutional neural networks is designed to learn the representation of the graph, which is used to classify graph nodes and simulate the relation prediction. The proposed model realizes the automatic nodes classification and network anomaly cause tracing. It is also applied to the public datasets in an unsupervised manner. Finally, the results show that the classification accuracy of the proposed model is better than the existing unsupervised graph neural network models, such as VGAE and ARVGE.