Adjusted Rand Index Measure

An Adjusted Rand Index Measure is an external clustering evaluation measure that measures clustering similarity between two clustering partitions adjusted for chance agreement.

• AKA: ARI Measure, Adjusted Rand Measure, Chance-Corrected Rand Index.
• Context:
  • It can typically compare unsupervised clustering output with ground truth labels.
  • It can typically assess clustering agreement through pairwise comparisons.
  • It can typically normalize Rand Index values using expected value correction.
  • It can often evaluate legal document clustering tasks for quality assessment.
  • It can often complement Normalized Mutual Information Measure in clustering evaluation.
  • It can range from being a Negative Adjusted Rand Index Measure to being a Perfect Adjusted Rand Index Measure, depending on its agreement level.
  • It can range from being a Binary Adjusted Rand Index Measure to being a Multi-Class Adjusted Rand Index Measure, depending on its cluster count.
  • It can range from being a Balanced Adjusted Rand Index Measure to being a Imbalanced Adjusted Rand Index Measure, depending on its cluster distribution.
  • It can range from being a Small-Dataset Adjusted Rand Index Measure to being a Large-Dataset Adjusted Rand Index Measure, depending on its data scale.
  • ...
• Examples:
  • Legal Clustering Evaluations, such as:
    • ACORD Clause Clustering ARI with 0.051 score.
    • CUAD Document Clustering ARI for contract grouping.
  • Benchmark Evaluations, such as:
    • Text Clustering ARI for document similarity.
    • Topic Model ARI for theme discovery.
  • ...
• Counter-Examples:
  • Silhouette Coefficient Measure, which is an internal measure.
  • Raw Rand Index, which lacks chance correction.
  • Accuracy Measure, which applies to supervised tasks.
• See: External Clustering Evaluation Measure, Clustering Evaluation Measure, Rand Index, Normalized Mutual Information Measure, Clustering Task, Cluster Purity Measure, F-Score Measure, Contingency Table, Pairwise Comparison.

---