Metadata Clustering
Cluster documents by attribute values for category discovery and organization
Why do anything?
Documents with similar metadata should be grouped. Manual categorization doesn't scale.
Why now?
Attribute patterns reveal natural groupings in data.
Why this feature?
Cluster documents by payload field combinations with configurable algorithms and LLM labeling.
How It Works
Metadata clustering groups documents by attribute similarity.
1
Field Extraction
Extract clustering fields from payloads
2
Feature Encoding
Encode categorical and numeric fields
3
Clustering
Apply algorithm to find groups
4
Labeling
LLM generates descriptive cluster names
Why This Approach
Attribute clustering complements vector clustering. Works when embeddings are unavailable.
Where This Is Used
Integration
cluster = client.clusters.create(type="metadata", fields=["category", "brand"])