Skip to main content

PS9594: Computational Text Analysis

“Computational Text Analysis”

 

Welcome to the site for the course PS9594A: “Computational Text Analysis” at Western University, taught by Sebastián Vallejo Vera. Each week, you will find the lecture slides, lecture code, exercises, and other code for the corresponding topic.

Before you start, don’t forget to read the Syllabus and check Perusall for the course readings. This site will be corrected and updated throughout the semester.

Course Overview

One of the most abundant sources of data available to social and political scientists today is text. Recent advances in Natural Language Processing (NLP) have spearheaded a text-as-data revolution, leading social scientists to seek new ways to analyze text data at scale. In this course, we will learn the intuition behind–and how to implement–different computational methods to process, analyze, and classify text. The course will cover Bag-of-Words (BoW) approaches, unsupervised methods, supervised and semi-supervised methods, and LLM-based approaches, as well as how to interpret the results obtained from applying these methods.

Readings, Assignments, and Final Exam

  • You can check a complete week-by-week reading list here. Note that the complete texts can be found in the course’s Perusall page.

  • You can check the schedule for the assignments, as well as instructions on presentation and submission, here.

  • You can check the instructions for the replication exercise here.

  • You can check the instructions for the final paper exercise here.

Software and Packages

For the first part of this course (Weeks 1 - 5), we will be mainly using R. For the second part of the this course (Weeks 6 - 11), we will use a combination of R and Python. I will assume that you are familiar with R language, RStudio, and R packages. In R, these are the main packages you will need to have installed:

  • tidyverse (we will be piping)
  • tidylog (helps keep track of what your are piping)
  • tidytext (great for working with text)
  • quanteda (stands for “Quantitative Analysis of Textual Data”)
    • quanteda.textstats (to obtain stats from our dfm)
    • quanteda.textplots (to obtain plots from our dfm stats)
    • quanteda.dictionaries (to use dictionaries with quanteda)
  • gutenbergr (to download texts from Project Gutenberg)
  • wesanderson (to make things pretty)
  • stm (to run Structural Topic Models)

Acknowledgments

The organization of the first part of this course (Weeks 1 - 5) and the format of the assignments are borrowed from many sources, among them Christopher Barrie’s excellent course on “Computational Text Analysis”, a syllabus from Tiago Ventura, and Grimmer, Roberts, and Stewart’s excellent book, “Text as data: A new framework for machine learning and the social sciences”. The code used throughout the course is a patchwork of my own code, but my own code borrows heavily from the internet (but that’s true for all code). I try my best to give credit to the original authors of the code (when and if possible).