Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Overlapping Community Detection using Dynamic Dilated Aggregation in Deep Residual GCN

Md Nurul Muttakin, Md Iqbal Hossain, Md Saidur Rahman

TL;DR

A deep dynamic residual graph convolutional network (DynaResGCN) based on the novel dynamic dilated aggregation mechanisms and a unified end-to-end encoder-decoder-based framework to detect overlapping communities in networks is designed.

Abstract

Overlapping community detection is a key problem in graph mining. Some research has considered applying graph convolutional networks (GCN) to tackle the problem. However, it is still challenging to incorporate deep graph convolutional networks in the case of general irregular graphs. In this study, we design a deep dynamic residual graph convolutional network (DynaResGCN) based on our novel dynamic dilated aggregation mechanisms and a unified end-to-end encoder-decoder-based framework to detect overlapping communities in networks. The deep DynaResGCN model is used as the encoder, whereas we incorporate the Bernoulli-Poisson (BP) model as the decoder. Consequently, we apply our overlapping community detection framework in a research topics dataset without having ground truth, a set of networks from Facebook having a reliable (hand-labeled) ground truth, and in a set of very large co-authorship networks having empirical (not hand-labeled) ground truth. Our experimentation on these datasets shows significantly superior performance over many state-of-the-art methods for the detection of overlapping communities in networks.

Overlapping Community Detection using Dynamic Dilated Aggregation in Deep Residual GCN

TL;DR

A deep dynamic residual graph convolutional network (DynaResGCN) based on the novel dynamic dilated aggregation mechanisms and a unified end-to-end encoder-decoder-based framework to detect overlapping communities in networks is designed.

Abstract

Overlapping community detection is a key problem in graph mining. Some research has considered applying graph convolutional networks (GCN) to tackle the problem. However, it is still challenging to incorporate deep graph convolutional networks in the case of general irregular graphs. In this study, we design a deep dynamic residual graph convolutional network (DynaResGCN) based on our novel dynamic dilated aggregation mechanisms and a unified end-to-end encoder-decoder-based framework to detect overlapping communities in networks. The deep DynaResGCN model is used as the encoder, whereas we incorporate the Bernoulli-Poisson (BP) model as the decoder. Consequently, we apply our overlapping community detection framework in a research topics dataset without having ground truth, a set of networks from Facebook having a reliable (hand-labeled) ground truth, and in a set of very large co-authorship networks having empirical (not hand-labeled) ground truth. Our experimentation on these datasets shows significantly superior performance over many state-of-the-art methods for the detection of overlapping communities in networks.
Paper Structure (30 sections, 14 equations, 7 figures, 8 tables, 2 algorithms)

This paper contains 30 sections, 14 equations, 7 figures, 8 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Notations and Terminologies
  4. Overlapping Community Detection Methods
  5. SLPA
  6. BigClam
  7. DEMON
  8. MNMF
  9. DANMF
  10. NOCD
  11. Our Methodology
  12. Deep Dynamic Residual GCN Encoder
  13. Residual Connection
  14. Dilated Aggregation
  15. Dynamicity of Edges
  16. ...and 15 more sections

Figures (7)

  • Figure 1: The overlapping community detection model based on the deep DynaResGCN encoder and the Bernoulli-Poisson decoder.
  • Figure 2: The architecture of the deep DynaResGCN encoder.
  • Figure 3: (a) A graph (network) (b) in which a node (red) is considered, (c) first-hop neighbors and second-hop neighbors are indicated in yellow and green, respectively, (d) graph augmentation up to second-hop neighbors, (e) dilated neighborhood at one layer (random), and (f) dilated neighborhood at another layer (random)
  • Figure 4: Two overlapping communities in a small portion of the topics network.
  • Figure 5: Overlapping tendency of selected topics from the topics dataset. The darker the cell, the higher the overlap.
  • ...and 2 more figures