Synthetic network generation with realistic cluster connectivity

Anne, Lahari

Permalink

https://hdl.handle.net/2142/129181 Copy

Description

Title

Synthetic network generation with realistic cluster connectivity

Author(s)

Anne, Lahari

Issue Date

2025-03-24

Director of Research (if dissertation) or Advisor (if thesis)

Chacko, George

Department of Study

Siebel School Comp & Data Sci

Discipline

Computer Science

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• synthetic networks
• community detection

Abstract

Evaluating the effectiveness of community detection methods is challenging due to the scarcity of real-world networks with known ground-truth communities. To address this, synthetic networks with predefined communities serve as valuable benchmarks. Among various synthetic network generators, Stochastic Block Models (SBMs) are widely used as they can approximate real-world network properties when provided with input parameters derived from real-world networks. However, SBMs often generate disconnected clusters, even when the input clustering exhibits fully connected communities, leading to structural inconsistencies that may affect the accuracy of performance evaluations for community detection algorithms. In this study, we introduce the REalistic Cluster Connectivity Simulator (RECCS), a post-processing framework designed to enhance SBM-generated networks by improving their fit to the cluster edge connectivity observed in real-world networks. RECCS modifies the synthetic network structure to better capture intra-cluster connectivity while preserving other essential network and clustering properties. This approach is evaluated on large-scale real-world networks containing up to 13.9 million nodes. The results show that RECCS generally improves the alignment of synthetic networks with empirical cluster connectivity, with some minimal trade-offs observed in other network properties. These findings suggest that RECCS offers a useful solution for generating synthetic benchmarks that more closely reflect real-world community structures, highlighting both its potential and limitations for community detection research.

Graduation Semester

2025-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129181

Copyright and License Information

Copyright 2025 Lahari Anne

Owning Collections