POL42050

Computational text analysis has become increasingly popular in political science in recent years. With the vast availability of text data on the web, political scientists increasingly view quantitative text analysis (or “text as data”) as a valuable approach for studying various forms of social and political behaviour.

This module introduces political science students to the quantitative analysis of textual data. It covers the theoretical foundations, practical applications, and technical implementations of these methods using the R statistical programming language. The module also explores advanced techniques, including word embeddings, speech transcription, and machine translation. Additionally, students will engage with the Hugging Face Python infrastructure, a cutting-edge resource for implementing transformer models and other state-of-the-art natural language processing methods.

Each session integrates lectures with practical, hands-on exercises to apply these methods to political texts. These exercises address practical challenges at each stage of the research process. Most of the methods follow a three-step framework: first, identifying texts and units of analysis; second, extracting measurable features from these texts and converting them into a quantitative feature matrix; and third, analysing this matrix using statistical techniques such as dictionary construction and application, scaling models, and topic models. Students will learn to apply these steps to various types of texts.

Building on this foundational framework, students will also gain hands-on experience with advanced techniques such as word embeddings, transformer models, and generative AI. These approaches will provide insights into the latest developments in text analysis and their applications to political science research.