Two-Level Clustering
Secondary clusters form behind a specific primary
- No fixed number of secondaries for a given primary
- The secondary membership threshold allows for gross tuning of cluster coherence, and hence precision
- Additional declaration threshold allows for control of document declaration similarity level distinct from within-primary clustering
Most learning occurs at this level
- When a secondary exceeds a similarity threshold with its primary, it declares its current document, updates the example vocabulary and is then colored appropriately