thai-taxonomy-expert

Designs and manages Thai product taxonomy hierarchies with hybrid keyword and embedding classification.

Updated Jul 12, 2025
One-click install
npx skills add https://github.com/kpayakv2/check-products --skill thai-taxonomy-expert-kpayakv2
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: thai-taxonomy-expert
Source: https://github.com/kpayakv2/check-products/tree/main/.agents/skills/thai-taxonomy-expert
Command: npx skills add https://github.com/kpayakv2/check-products --skill thai-taxonomy-expert-kpayakv2

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires httpx.

What problem does it solve? Managing a Thai product category hierarchy is error-prone: nodes get created without enough keywords, embeddings are forgotten after inserts, and edits to taxonomy_nodes silently break classification for thousands of products. This Skill provides the schema knowledge, design rules, and safe workflows to design, extend, and modify a Thai product taxonomy without regressing classification accuracy. ## Core Features & Use Cases - Taxonomy Hierarchy Design: Enforces a 3-level structure (L1 main, L2 sub, L3 specific) with rules such as minimum 3 Thai keywords per node and no ambiguous catch-all categories. - Hybrid Classification Guidance: Documents the 60% keyword / 40% embedding (384-dim vector, cosine distance) algorithm and how to tune keyword_rules priorities. - Safe Modification Workflow: Provides blast-radius SQL checks (product counts, child nodes) and backup thresholds before editing or deleting nodes. - Use Case: You need to add a new "นมสด" (fresh milk) category under the dairy parent node. The Skill walks you through checking the parent hierarchy, defining Thai keywords and brand aliases, inserting the node, generating its embedding via the local FastAPI server, and verifying classification with a test product name. ## Quick Start Ask the assistant to add a new Thai product category to the taxonomy, and it will check the parent hierarchy, define keywords, insert the node, generate its embedding, and test classification.

Frequently Asked Questions about thai-taxonomy-expert

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I add a new category to a Thai product taxonomy?▼

First query taxonomy_nodes to find the correct parent, then define at least 3 Thai keywords covering product names, brands, and units. Insert the node, generate its 384-dim embedding via the local FastAPI /api/embed endpoint, and test classification with a sample product name.

How does hybrid keyword and embedding product classification work?▼

The classifier combines keyword matching (60% weight) against keyword_rules and taxonomy_nodes.keywords with vector similarity (40% weight) using cosine distance on 384-dimensional embeddings. The combined score determines the assigned category.

What should I check before editing or deleting a taxonomy node?▼

Run a blast-radius check: count products classified into the node and list its child nodes. If the impact exceeds 10% of total data, back up the database before making changes.

Why does Thai product classification fail after adding a new category?▼

Common causes are missing embeddings (the node was inserted without calling the embedding API), fewer than 3 keywords, or overly broad node names that create ambiguous matches. Always generate the embedding and test classification after insertion.

What are the limitations of keyword-based Thai text classification?▼

Keyword matching alone struggles with Thai word segmentation and unseen aliases, which is why the system blends it with multilingual sentence embeddings. Pure keyword rules also require manual maintenance as new brands and product terms appear.