Measuring Task-Level Technological Exposure: A Language Model Approach

This repository contains the public dataset for "Measuring Task-Level Technological Exposure: A Language Model Approach" (click here for the latest draft). See github.com/armouton/task-tech-exposure for the accompanying Python package and installation instructions. It is recommended that you download data through the Python package rather than from this webpage, as this ensures the necessary file structure. The package is in beta, so please report any issues! All data are drawn from public sources: the Department of Labor's ONET (versions 6.0-30.0), the Bureau of Labor Statistics' OEWS (annual releases, 2001-2025), and USPTO patent application publications (2001-2025). The included Sentence-BERT models are fine-tuned versions of Sentence Transformers' mpnet-base-v2. Note that patent files are organized by publication date, which typically occurs 18 months after the date of filing but may, in some cases, involve a shorter or a longer time lag. For best results, you should download patent data for any year(s) of interest and also for all subsequent years.