Dynamic taxonomy generation for future skills identification using a named entity recognition and relation extraction...

Area of Science:

• Computational Linguistics
• Labor Economics
• Information Science

Background:

• The labor market is rapidly evolving, creating a gap between current Knowledge, Skills, and Abilities (KSAs) and future occupational needs.
• Organizations like the World Economic Forum and OECD highlight the necessity for dynamic skill identification to adapt to these changes.

Purpose of the Study:

• To develop a novel system for constructing a dynamic taxonomy of skills.
• To utilize Natural Language Processing (NLP) techniques, including Named Entity Recognition (NER) and Relation Extraction (RE), for identifying and predicting future skills.
• To bridge the gap between current workforce competencies and future demands, supporting educational and professional development.

Main Methods:

• An NLP-based architecture was developed, incorporating text preprocessing, NER, and RE models.
• NER models identified and categorized KSAs and occupations from labor market reports.
• RE models established semantic relationships between identified entities, trained on over 1,700 annotated documents.

Main Results:

• The NER model achieved a micro-averaged F1-score of 65.38% for entity recognition.
• The RE model achieved a micro-F1 score of 82.2% for relationship classification.
• The generated taxonomy successfully identified emerging skills and occupations, demonstrating dynamic adaptability to labor market shifts.

Conclusions:

• The dynamic taxonomy provides real-time updates on competencies and predicts emerging skill trends, serving as a valuable workforce planning tool.
• While NER showed strong recall, precision improvements are needed; future work will expand the corpus and refine models.
• The system offers insights into future workforce requirements and can be applied across multiple sectors.