Cascading Clustering (E-commerce)

Cascading clustering is a multi-level clustering approach designed for e-commerce sites, where categories, subcategories, and products form a cascade of clusters with different granularity. Instead of single-level grouping of 500 keywords, you cluster cascadingly: first broad categories (shoes, bags, jackets), then subcategories (athletic shoes, dress shoes), and finally products (Nike Air Max 90, Adidas Superstar).

Each level requires different k in K-means and different Silhouette Score thresholds. Unlike standard clustering, cascading preserves e-commerce navigation hierarchy — breadcrumbs map directly to clustering levels. The approach is analogous to the ROOT > SEED > NODE hierarchy from knowledge graphs.

In practice for e-commerce with 10,000+ products, start from category clustering (top-level), then iteratively dive into subcategories.