hierarchical-taxonomy-clustering

What problem does it solve?

Given disparate category paths from multiple sources (for example Electronics > Computers > Laptops), this approach builds a single, unified taxonomy that groups similar paths, assigns meaningful names, and outputs a clean, fixed-depth hierarchy (typically five levels) for analysis and cross-platform comparisons.

Core Features & Use Cases

• Hierarchical weighting and embedding-based representation of category paths to capture semantic similarity across sources.
• Recursive clustering with cosine distance to form a coherent, multi-level taxonomy (levels 1–5 by default).
• Intelligent naming by combining weighted terms and lemmatization to generate human-readable category labels.
• Quality control to remove duplicates and prevent ancestor path collisions, enabling stable taxonomy across datasets.

Use cases include harmonizing vendor catalogs, enabling cross-platform analytics, and powering downstream search or recommendation tasks.

Quick Start

Run the full 4-step pipeline to process multiple sources and export unified taxonomy CSVs.