The paper introduces V-Measure, an external entropy-based cluster evaluation measure designed to address several issues with existing cluster evaluation methods. V-Measure combines homogeneity and completeness, two complementary criteria for evaluating clustering solutions, using a harmonic mean. This approach ensures that the measure is independent of the clustering algorithm, dataset size, number of classes, and clusters, and it avoids the "problem of matching" by evaluating the entire membership of each cluster. The paper compares V-Measure with other popular measures such as Purity, Entropy, F-Measure, Rand Index, and variation of information (VI), demonstrating that V-Measure satisfies several desirable properties, including being n-invariant and k-invariant. The authors also evaluate V-Measure on document clustering and pitch accent type clustering tasks, showing its effectiveness in comparing clustering solutions across different domains.The paper introduces V-Measure, an external entropy-based cluster evaluation measure designed to address several issues with existing cluster evaluation methods. V-Measure combines homogeneity and completeness, two complementary criteria for evaluating clustering solutions, using a harmonic mean. This approach ensures that the measure is independent of the clustering algorithm, dataset size, number of classes, and clusters, and it avoids the "problem of matching" by evaluating the entire membership of each cluster. The paper compares V-Measure with other popular measures such as Purity, Entropy, F-Measure, Rand Index, and variation of information (VI), demonstrating that V-Measure satisfies several desirable properties, including being n-invariant and k-invariant. The authors also evaluate V-Measure on document clustering and pitch accent type clustering tasks, showing its effectiveness in comparing clustering solutions across different domains.