Occupational models from 42 million unstructured job postings.

Area of Science:

• Labor economics
• Computational linguistics
• Public health informatics

Background:

• Accurate job classification is crucial for economic and public health research, and for workforce development.
• Existing methods for assigning occupation codes to job postings can be labor-intensive and imprecise.
• A data-driven approach is needed to model the complex relationships between job titles, skills, and occupational categories.

Purpose of the Study:

• To develop and present a data resource and computational models for empirically linking unstructured job postings to standardized occupation codes.
• To create an open-source Python package (sockit) that facilitates the automated assignment of occupation codes and skill extraction from job data.
• To enhance the analysis of labor market dynamics and improve job matching processes.

Main Methods:

• Utilized natural language processing (NLP) techniques on over 42 million U.S. job postings from 2019-2021.
• Developed probabilistic models to associate job titles with occupation codes and job descriptions with relevant skills and occupation codes.
• Leveraged the Standardized Occupation Coding for Computer-assisted Epidemiological Research (SOCR) method for initial occupation code estimation.

Main Results:

• Created a comprehensive dataset modeling associations between occupation codes, skills, job titles, and descriptions.
• The sockit Python package can accurately assign occupation codes to job titles and postings.
• The package enables skill parsing from job postings and resumes, and estimates occupational similarity.

Conclusions:

• The developed NLP models and sockit package provide a scalable and accurate method for job classification and skill analysis.
• This resource can significantly improve the efficiency of labor market research, public health studies, and job matching platforms.
• Open-source availability of the models and tools promotes wider adoption and further research in occupational data science.