V-measure
V-measure (Homogeneity and Completeness Harmonic Mean) · Also known as: V-measure score, homogeneity completeness V-measure
V-measure, introduced by Rosenberg and Hirschberg in 2007, is an external clustering evaluation metric based on the harmonic mean of homogeneity and completeness. It measures whether clusters contain only points from a single true class (homogeneity) and whether all points from a true class are assigned to the same cluster (completeness). Values range from 0 to 1.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use V-measure when you want a balanced external clustering metric that decomposes into interpretable components. It is particularly useful when you want to diagnose whether clustering performance is limited by purity or by coverage. V-measure works well on datasets with well-defined class structure and is comparable across studies.
Strengths & limitations
- Decomposable into homogeneity and completeness; provides diagnostic information
- Ranges from 0 to 1 for intuitive interpretation
- Balances precision-like (homogeneity) and recall-like (completeness) properties
- Flexible weighting through beta parameter for domain-specific priorities
- Symmetric and interpretable as an F-measure for clustering
- Requires ground truth labels; not suitable for unsupervised evaluation
- Assumes categorical labels; less suitable for overlapping or hierarchical clusters
- Can be difficult to interpret without understanding conditional entropy
- Sensitive to the number of classes and class imbalance
Frequently asked
What does high homogeneity but low completeness indicate?
This pattern suggests that your clusters are pure—each cluster contains mostly points from one true class—but that true classes are fragmented across multiple clusters. You may need to merge clusters or reconsider your clustering approach to improve completeness.
How do homogeneity and completeness relate to precision and recall?
Homogeneity is conceptually similar to precision: it measures purity within clusters. Completeness is similar to recall: it measures the recovery of true classes. V-measure is the harmonic mean of these two, analogous to the F1 score in classification.
Can I adjust V-measure to emphasize one criterion?
Yes, the beta parameter in V-measure controls the weighting. beta > 1 emphasizes completeness; beta < 1 emphasizes homogeneity. Set beta based on your application's priorities.
How is V-measure different from Normalized Mutual Information?
Both are information-theoretic metrics, but V-measure explicitly decomposes clustering quality into homogeneity and completeness, providing diagnostic information. NMI measures shared information without explicit decomposition. Choose V-measure for interpretability and NMI for information-theoretic properties.
Sources
- Rosenberg, A., & Hirschberg, J. (2007). V-measure: A conditional entropy-based external cluster evaluation measure. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (pp. 410-420). link ↗
How to cite this page
ScholarGate. (2026, June 3). V-measure (Homogeneity and Completeness Harmonic Mean). ScholarGate. https://scholargate.app/en/model-evaluation/v-measure
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Adjusted Rand IndexModel Evaluation↔ compare
- Davies-Bouldin IndexModel Evaluation↔ compare
- Fowlkes-Mallows IndexModel Evaluation↔ compare
- Normalized Mutual InformationModel Evaluation↔ compare
- Silhouette ScoreModel Evaluation↔ compare