Reference · Glossary

Clustering

Last updated

Exploratory analysis: segment users, theme open-ended survey answers, or bucket documents before labeling.

#When to use

Exploratory analysis: segment users, theme open-ended survey answers, or bucket documents before labeling.

#When not to

When you already have reliable labels and need a classifier. Clusters are hypotheses, not ground truth.

#Quality checklist

  • Choose distance metric that matches your embedding space
  • Inspect exemplar points per cluster
  • Check stability when you re-run with new seeds
  • Name clusters from evidence, not wishful thinking

#Example

Embed 2,000 support tickets → k-means (k=12)
Label each cluster by reading 10 random tickets
Promote stable themes into a taxonomy for classification

#Learn next

  • Lesson: `clustering-basics`
  • Related: embedding