Using Natural Language Processing to Facilitate Social Science Research

Abstract

Recent advances in natural language processing in the form of large language models (e.g., ChatGPT, GPT-3, BERT) have created new opportunities for social science research. While some of these models are proprietary and not easily accessible to researchers, others are publicly available through open source repositories such as HuggingFace. Key to these new language models is their ability to capture semantic meaning in texts, which means social scientists can leverage them to identify themes in large corpora of text that were previously unwieldy to analyze. In this workshop, we will review methods to harness these models to glean information from a range of sources including interviews, open-ended surveys, and web-scraped data.