Silhouette Coefficient Measure

A Silhouette Coefficient Measure is an internal clustering evaluation measure that quantifies cluster cohesion and cluster separation through distance measurements.

• AKA: Silhouette Score, Silhouette Width, Silhouette Index.
• Context:
  • It can typically evaluate clustering quality without external labels.
  • It can typically measure intra-cluster distances for cohesion assessment.
  • It can typically calculate inter-cluster distances for separation evaluation.
  • It can often validate unsupervised legal document clustering tasks through quality scores.
  • It can often guide optimal cluster number selection via score maximization.
  • It can range from being a Negative Silhouette Coefficient Measure to being a Perfect Silhouette Coefficient Measure, depending on its score value.
  • It can range from being a Euclidean Silhouette Coefficient Measure to being a Cosine Silhouette Coefficient Measure, depending on its distance measure.
  • It can range from being an Individual Silhouette Coefficient Measure to being an Average Silhouette Coefficient Measure, depending on its aggregation level.
  • It can range from being a Dense-Cluster Silhouette Coefficient Measure to being a Sparse-Cluster Silhouette Coefficient Measure, depending on its cluster density.
  • ...
• Examples:
  • Legal Document Clustering Evaluations, such as:
    • ACORD Dataset Silhouette Score with 0.927 value.
    • CUAD Contract Silhouette Analysis for quality assessment.
  • Clustering Algorithm Comparisons, such as:
    • K-Means Silhouette Score for partition quality.
    • Hierarchical Clustering Silhouette for dendrogram cut.
  • ...
• Counter-Examples:
  • Adjusted Rand Index Measure, which is an external measure.
  • Davies-Bouldin Index, which uses different calculation.
  • Supervised Accuracy Measure, which requires labeled data.
• See: Internal Clustering Evaluation Measure, Clustering Evaluation Measure, Calinski-Harabasz Score Measure, Davies-Bouldin Index, Dunn Index, Cluster Cohesion, Cluster Separation, Distance Measure, Clustering Task.

---