Enhanced incremental dynamic community detection for improving the stability of community structure in network analysis
Network analysis plays a crucial role in detecting communities within complex networks, which are prevalent across diverse domains and exhibit intricate structures and interactions. Dynamic community detection (DCD) is essential for understanding evolving complex networks, where vertices and edges c...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | en |
| Published: |
2026
|
| Subjects: | |
| Online Access: | http://eprints.utem.edu.my/id/eprint/29654/1/Enhanced%20incremental%20dynamic%20community%20detection%20for%20improving%20the%20stability%20of%20community%20structure%20in%20network%20analysis.pdf http://eprints.utem.edu.my/id/eprint/29654/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Network analysis plays a crucial role in detecting communities within complex networks, which are prevalent across diverse domains and exhibit intricate structures and interactions. Dynamic community detection (DCD) is essential for understanding evolving complex networks, where vertices and edges continuously change over time. However, maintaining stability and continuity in dynamic networks remains a significant challenge, as communities experience birth, death, splitting, or merging. Existing techniques often struggle with these evolving dynamics, causing inconsistent tracking and reduced stability. To address these limitations, this thesis proposes an enhanced incremental dynamic community detection called Dynamic Community Detection based on the Bird Flock Effect (DCDBFE). This technique was inspired by the behavior of bird flocks and utilizes the principles of separation, alignment, and cohesion. The proposed technique incorporates a Resource Allocation similarity measure and third-level module attraction function to dynamically update community structures. This incremental approach effectively captures temporal changes and ensures continuous community structure tracking, unlike traditional static models. To ensure rigorous evaluation, synthetic networks were generated using an extended Lancichinetti- Fortunato-Radicchi benchmark, specifically designed in three scales, low, medium, and large with controlled parameters for mixing coefficient, edge density, and probability of vertex switching. These network datasets enabled systematic assessment of module stability across different network complexities. The experimental results were further validated using real-world network datasets to confirm the stability of the proposed technique. Stability was quantified based on inter-snapshot similarity, using metrics such as the Normalized Mutual Information and Adjusted Rand Index. Statistical validation was conducted using the Friedman test with a 95% confidence level to confirm the significance of performance improvements among competing techniques. The results demonstrated that DCDBFE achieved up to 15-25% higher stability compared to state-of-the-art DCD techniques. This research contributes to the field of dynamic community detection by providing a stable solution through an innovative methodology inspired by natural phenomena. Future work will extend DCDBFE to heterogeneous and weighted networks, and integrate deep learning- based adaptive similarity functions to further enhance temporal prediction and real-time community detection analysis. |
|---|