Skip to main content
Search

# Agentic Taxonomy Construction

1 related articles

Subscribe to updates
  1. arXiv

    Four agentic runs produced customer-feedback taxonomies that passed every generic check while 97.7%–100% of leaf names restated an ancestor and branch leakage reached 76.4%

    Building six taxonomies over two proprietary customer-feedback corpora (1,940 and 5,000 records) — one production reference plus two repeated runs of the same agentic harness per corpus — the authors found all six pass every generic naming and structure check, yet at least 97.7% of leaf names in every generated tree merely restate an ancestor's name and one tree routes three of four records into multiple top-level categories; they introduce two judge-free whole-tree metric families, structural discriminators and team partitionability.

You’re all caught up