Search:
Match:
1 results

Analysis

This paper investigates the effectiveness of the silhouette score, a common metric for evaluating clustering quality, specifically within the context of network community detection. It addresses a gap in understanding how well this score performs in various network scenarios (unweighted, weighted, fully connected) and under different conditions (network size, separation strength, community size imbalance). The study's value lies in providing practical guidance for researchers and practitioners using the silhouette score for network clustering, clarifying its limitations and strengths.
Reference

The silhouette score accurately identifies the true number of communities when clusters are well separated and balanced, but it tends to underestimate under strong imbalance or weak separation and to overestimate in sparse networks.