Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Community Detection of Complex Network Based on Graph Convolution Iterative Algorithm

Jiaqi Yao, Lewis Mitchell

TL;DR

The paper tackles non-overlapping community detection in complex networks and the high cost of training-weighted GCNNs. It introduces Graph Convolution Iteration Algorithm (GCIS), a weight-free, iterative convolution framework that builds a node attribute matrix from distances to randomly sampled candidate centers and optimizes partitioning via modularity $Q$ without learning weights. A novel proximity-based initialization, normalized adjacency convolution, and max-affiliation partitioning enable accurate, scalable detection validated across random and real networks. The work highlights a practical alternative to training-based GCNNs and sets the stage for extending the approach to overlapping communities and broader network types.

Abstract

Community detection can reveal the underlying structure and patterns of complex networks, identify sets of nodes with specific functions or similar characteristics, and study the evolution process and development trends of networks. Despite the myriad community detection methods that have been proposed, researchers continue to strive for ways to enhance the accuracy and efficiency of these methods. Graph convolutional neural networks can continuously aggregate the features of multiple neighboring nodes and have become an important tool in many fields. In view of this, this paper proposes a community detection method for complex networks based on graph convolution iteration algorithm. Firstly, the candidate community centers are determined by random sampling and the node attribute matrix is obtained based on the distances of nodes to community centers. Next, the graph convolution operation is implemented to obtain the convolutional node attribute matrix. Then, community partitioning method according to the convolutional node attribute matrix is presented and the effectiveness of community partitioning is measured through modularity. The method proposed in this paper is applied into to multiple random and real-world networks. The comparison results with some baseline methods demonstrate its effectiveness.

Community Detection of Complex Network Based on Graph Convolution Iterative Algorithm

TL;DR

The paper tackles non-overlapping community detection in complex networks and the high cost of training-weighted GCNNs. It introduces Graph Convolution Iteration Algorithm (GCIS), a weight-free, iterative convolution framework that builds a node attribute matrix from distances to randomly sampled candidate centers and optimizes partitioning via modularity without learning weights. A novel proximity-based initialization, normalized adjacency convolution, and max-affiliation partitioning enable accurate, scalable detection validated across random and real networks. The work highlights a practical alternative to training-based GCNNs and sets the stage for extending the approach to overlapping communities and broader network types.

Abstract

Community detection can reveal the underlying structure and patterns of complex networks, identify sets of nodes with specific functions or similar characteristics, and study the evolution process and development trends of networks. Despite the myriad community detection methods that have been proposed, researchers continue to strive for ways to enhance the accuracy and efficiency of these methods. Graph convolutional neural networks can continuously aggregate the features of multiple neighboring nodes and have become an important tool in many fields. In view of this, this paper proposes a community detection method for complex networks based on graph convolution iteration algorithm. Firstly, the candidate community centers are determined by random sampling and the node attribute matrix is obtained based on the distances of nodes to community centers. Next, the graph convolution operation is implemented to obtain the convolutional node attribute matrix. Then, community partitioning method according to the convolutional node attribute matrix is presented and the effectiveness of community partitioning is measured through modularity. The method proposed in this paper is applied into to multiple random and real-world networks. The comparison results with some baseline methods demonstrate its effectiveness.

Paper Structure

This paper contains 27 sections, 20 equations, 11 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Graph Partitioning Method
  4. Similarity Based Method
  5. Label Propagation Algorithm
  6. Deep Learning Based Method
  7. Preliminary
  8. Problem Description
  9. Evaluation Indicator
  10. Traditional Graph Convolutional Neural Network
  11. Community Detection Method Based on Graph convolution Iteration Algorithm
  12. Overall Process
  13. Initialization of Node Attribute Matrix
  14. Graph Convolution Operation
  15. Community Partitioning
  16. ...and 12 more sections

Figures (11)

  • Figure 1: Changing curves of different indicators for the example network.
  • Figure 2: Information of First Group of Random Networks.
  • Figure 3: Information of Second Group of Random Networks.
  • Figure 4: Changing curves of different indicators for the example network.
  • Figure 5: Modularities obtained by different methods in the first set of random networks.
  • ...and 6 more figures