Highway algorithm improves overlapping community detection by 6.9%
New algorithm exploit sparse backbones to cluster networks faster and more accurately.
Researchers Zihe Zhou and Samin Aref from arXiv have introduced Highway, a scalable overlapping community detection (OCD) algorithm that leverages the sparse backbone of a network to perform efficient community inference. Overlapping community detection allows nodes (e.g., users, proteins) to belong to multiple groups simultaneously, a common feature in real-world networks but one that often challenges existing algorithms in terms of balancing accuracy and scalability.
In a comprehensive evaluation using 728 Lancichinetti-Fortunato-Radicchio benchmark networks, Highway compared favorably against 10 existing OCD algorithms across five performance measures. It ranked first in overlapping normalized mutual information, achieving a 6.9% improvement over the strongest baseline, and scored second in all other four measures. The algorithm, now open-source and available in the CDlib library, offers practitioners a practical tool for tasks like drug discovery, market segmentation, and social network analysis, providing a strong accuracy-efficiency trade-off.
- Highway achieves a 6.9% improvement in overlapping normalized mutual information over the strongest baseline.
- Tested on 728 benchmark networks against 10 competing OCD algorithms across five metrics.
- Open-source and available in the CDlib library for immediate use by researchers and practitioners.
Why It Matters
Enables faster, more accurate detection of overlapping groups in large networks for real-world applications.