Building a Custom Taxonomy of AI Skills and Tasks from the Ground Up with Job Postings

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, quick

Summary

TaxonomyBuilder is introduced as a blueprint for systematically constructing custom, data-informed, and hierarchical taxonomies of AI skills in the workplace. This approach utilizes large language models (LLMs) for automated taxonomy construction, specifically investigating optimal data inclusion and exclusion strategies from two large-scale job postings corpora. The research demonstrates that filtering input data to TaxonomyBuilder yields superior domain-specific coverage compared to feeding unfiltered inputs directly into clustering and LLM-enhanced hierarchical taxonomy labeling tools. This finding suggests that a curated, "less data" approach can provide greater clarity and more effective mapping of complex domains like AI skills, despite the availability of high volumes of rapidly growing corpora.

Key takeaway

For NLP Engineers or Data Scientists building skill taxonomies from large text corpora, you should prioritize rigorous data filtering over simply feeding all available data to LLMs. Your taxonomy construction efforts will achieve better domain-specific coverage by curating inputs for tools like TaxonomyBuilder, rather than relying on unfiltered data for clustering and LLM-enhanced labeling. This approach ensures more accurate and relevant skill mapping for your specific workplace domain.

Key insights

Filtering input data for LLM-driven taxonomy construction significantly improves domain-specific coverage compared to using unfiltered data.

Principles

• Data quality trumps quantity in taxonomy construction.
• Strategic data filtering enhances domain specificity.
• LLMs benefit from curated inputs for complex mapping.

Method

Propose TaxonomyBuilder as a blueprint to evaluate data inclusion/exclusion for LLM-enhanced hierarchical taxonomy labeling using job postings corpora.

In practice

• Filter job posting data before LLM-based skill extraction.
• Prioritize domain-specific data for taxonomy inputs.
• Evaluate data configurations with TaxonomyBuilder blueprint.

Topics

• AI Skills Taxonomy
• Large Language Models
• Job Postings Analysis
• Data Filtering
• TaxonomyBuilder
• Workforce Analytics

Best for: Research Scientist, AI Scientist, NLP Engineer, Data Scientist

Related on AIssential

• Taxonomies, All the Way Down — Taxonomies, a pillar of library science, provide explicit, auditable structure that significantly enhances LLM accuracy and reliability.
• NLP: From Prototype to Production — Production NLP prioritizes specific business utility and maintainability over raw academic accuracy.
• Exploring a New Competency Modeling Process with Large Language Models — LLMs can transform expert-dependent competency modeling into a data-driven, evaluable analytical process.
• Knowledge Engineering for Search and Content: A Practical Guide — Structured understanding of content and user intent is crucial for effective LLM-powered search and content experiences.
• System Report for CCL25-Eval Task 5: New Dataset and LoRA-Fine-Tuned Qwen2.5 — Domain-specific datasets and LoRA fine-tuning significantly enhance LLM performance in specialized tasks like classical poetry analysis.
• Are LLMs safe? — LLM training data curation, especially via quality filters, embeds implicit biases that shape model behavior and capabilities.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.