Job Demand Estimation Using Text Embeddings of Patent Classification Codes and Occupational Data

This study presents a novel method for estimating job demand by analyzing the relationship between technological advancements and occupational requirements through patent data. By integrating textual data from the United States Patent and Trademark Office (USPTO) and the Occupational Information Network (O*NET), the research develops a framework using advanced text embedding techniques. Semantic similarity measures are applied to match Cooperative Patent Classification (CPC) code descriptions with job task descriptions, constructing a similarity matrix that quantifies the technological relevance of various occupations. The approach estimates annual job task demand based on patent filing trends, providing a data-driven perspective on labor market evolution. To enhance reliability, multiple text embedding models are assessed to determine the most effective one for measuring semantic similarities. Aggregated job task demand estimates are used to project overall job demand trends systematically. Validation is conducted by comparing predicted job demand metrics with empirical employment data, demonstrating a strong correlation between estimated and actual trends. The findings highlight the impact of technological progress on job market dynamics, identifying occupations with increasing or declining demand. The study also acknowledges limitations, such as the need for improved job-task matching and additional data integration, while discussing broader implications for labor market forecasting.