Semi-Supervised Label Propagation Community Detection on Graphs with Graph Neural Network

The Graph Neural Network is a fairly innovative concept which permits neural networks to function on random graphs. As unbounded problem structures are universal in real-world scenarios and can be best denoted by graphs, Graph Neural Networks suggests new exhilarating applications and further simplified latent for machine learning wholly, but also noteworthy enhancement of performance in a deep learning domain. Graph Neural Networks are variant of Graph convolution networks can function sprightly on graphs. One of the well-known tasks attempted with this new skill is graph partitioning. Important characteristic of community is to discover graph nodes are with same interests and keep them strongly connected to extract groups for numerous reasons. We demonstrate a semi-supervised learning on graph data for solving community detection. In a number of trials on graph partitions we proved that our framework outperforms traditional ones.

I. Introduction

In Modern days, Graphs gain a great deal of attention due to their ability to characterize the actuality in a manner that can be evaluated accurately. Graphs can be used to signify a modern dataset such as social platforms, protein-protein relations, molecule structures, friendship networks, web link data, knowledge graphs, etc. Even non-structured data like images and text can be exhibited as graphs. Graphs are data structures that miniature a set of nodes and edges. Graph analytics precisely focus on analyzing the relationship between nodes in a graph. The attention of graph analytics is on pairwise associations between nodes and structural features of the graph. Graph analytics also focuses on tasks like node classification, link prediction, graph clustering and visualization. Traditional clustering approaches is depended on the compactness of data elements. These are computationally expensive. Graph Neural Networks (GNN) are deep neural networks that can activate on graph domains. Owing to its good performance in real-world applications, GNN become an extensively useful graph analysis approach.