V-Measure: A Conditional Entropy-Based External Cluster Evaluation Measure

By Andrew Rosenberg, Julia Hirschberg. Conference on Empirical Methods in Natural Language Processing, 2007

1,885 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/5421dbcb7e14766eb3d951910ae8d7892d735a01

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Coincidence, Categorization, and Consolidation: Learning to Recognize Sounds with Minimal Supervision (2019, methodology): “Table 2 shows the context-window clustering performance in terms of V-Measure (VM) [37], a standard clustering metric, along with the corresponding label precision and recall resulting from the cluster-based labeling procedure.”
Analyzing Speaker Information in Self-Supervised Models to Improve Zero-Resource Speech Processing (2021, methodology): “By mapping each unit to the overlapping phone label in the forced alignment, we evaluate the clustering quality in terms of the adjusted rand index (ARI), adjusted mutual information (AMI) [26], homogeneity, and completeness [27].”
A Survey of Clustering Ensemble Algorithms (2011, background): “Some of them are the Class Entropy,(8) Normalized Mutual Information,(75) Utility Function,(64) Variation of Information(62) and V-measure.(70) .”
Discovering the Compositional Structure of Vector Representations with Role Learning Networks (2019, background): “The V-Measure [17] assesses the extent to which the clustering of the role vectors assigned by ROLE matches the ground truth role assignments.”
Measuring the perceptual availability of phonological features during language acquisition using unsupervised binary stochastic autoencoders (2019, methodology): “Table 1 reports learning outcomes using the information theoretic measures homogeneity (H), completeness (C), and V-measure (V) for unsupervised cluster evaluation (Rosenberg and Hirschberg, 2007).”
Dialect Diversity in Text Summarization on Twitter (2020, background): “V-measure, on the other hand, is to evaluate clustering tasks [63].”
Learning Syntactic Categories Using Paradigmatic Representations of Word Context (2012, background): “The V-measure (VM) (Rosenberg and Hirschberg, 2007) is an information theoretic metric that reports the harmonic mean of homogeneity (each cluster should contain only instances of a single class) and completeness (all instances of a class should be members of the same cluster).”
Learning Structural Node Embeddings via Diffusion Wavelets (2017, methodology): “We use agglomerative clustering (with single linkage) to cluster embeddings learned by each method and evaluate the clustering quality via: (1) homogeneity [28], conditional entropy of ground-truth structural roles given the predicted clustering; (2) completeness [28], a measure of how many nodes wi”
Deep Bayesian Unsupervised Lifelong Learning (2021, background): “Detailed definitions for these metrics can be found in [39].”
MOA: Massive Online Analysis, a Framework for Stream Classification and Clustering (2010, unknown): “Gamma [5] Rand statistic [35] C Index [24] Jaccard coefficient [19] Point-Biserial [30] Folkes and Mallow Index [19] Log Likelihood [22] Hubert Γ statistics [23] Dunn's Index [17] Minkowski score [12] Tau [37] Purity [44] Tau A [24] van Dongen criterion [16] Tau C [24] V-measure [38] Somer's Gamma [”
Large-scale audio event discovery in one million YouTube videos (2017, background): “This rules out popular single-value summary metrics like adjusted rand index and V-measure [20].”
Hands Holding Clues for Object Recognition in Teachable Machines (2019, methodology): “10, to allow for the comparison, we use V-measure [57] by comparing desirable properties of the two, such as consistency of their predictions given images of the same object, and ability to distinguish between two diferent objects.”
Automatic Spatially-Aware Fashion Concept Discovery (2017, methodology): “By jointly clustering the semantic and spatial representations of attributes, our concept discovery approach outperforms other methods by 0.03 in V-measure.”
Hierarchy-aware Loss Function on a Tree Structured Label Space for Audio Event Detection (2019, background): “• Homogeneity (H), Completeness (C) and V-measure (V) [28]: If the true classes are known, then perfect homogeneity (1.”
Efficient image dataset classification difficulty estimation for predicting deep-learning accuracy (2018, methodology): “We assess the following known scores: adjusted mutual information [27], adjusted rand index [28], completeness, homogeneity and the v-measure [29].”
Semi-supervised New Event Type Induction and Event Detection (2020, background): “V-Measure (Rosenberg and Hirschberg, 2007) is the weighted harmonic mean between homogeneity score and completeness score.”