Scientific Coordination

The course will provide insights into the concepts, challenges and opportunities associated with data so large that traditional research methods (like manual content analysis) can no longer be applied, and traditional inferential statistics start to lose their meaning. Participants are introduced to strategies and techniques for capturing and analyzing digital data in communication contexts using Python. The course offers hands-on instructions regarding the several stages of computer-aided content analysis. More precisely, students will be familiarized with pre-processing methods, analysis strategies and the visualization and presentation of findings. The focus will be in particular on Machine Learning techniques to analyze quantitative textual data, amongst which both deductive (e.g., supervised machine learning) and inductive (e.g., unsupervised machine learning) approaches will be discussed.

This is a beginner's course. Participants who are looking to learn about the latest developments in machine learning for textual data (such as large language models) should consider taking a different course, e.g., “From Embeddings to LLMs: Advanced Text Analysis in Python” (22-26 September). These techniques will be (briefly) discussed towards the end of the course, but the focus lies on the basics of natural language processing and classical machine learning in Python.

For additional details on the course and a day-to-day schedule, please download the full-length syllabus.

Organizational Structure of the Course

In the morning, we will have lectures, in which we will explain the topic of the day both from a theoretical-conceptual point of view as well as from a practical point of view (i.e., walking you through code examples). We may have small in-class exercises in between, if necessary.

In the afternoon, students work on larger exercises in which they implement the techniques we covered. We provide example datasets, but it is also possible (and strongly encouraged) to try to apply the techniques to own datasets. Participants can either opt to work on their own or try to solve problems together with one or multiple classmates. Lecturers will provide feedback on the (attempted) solutions of participants and also provide example solutions.

Target group

You will find the course useful if:

You are a social scientist who has the ambition to model quantitative textual data. Specifically, those who aim to describe, explain, or predict the content of large-scale textual data using computation techniques are likely to benefit from participating in this course.
Note that non-textual data, such as images or networks, are not at the center of this course. Techniques we cover are partly generalizable to such types of data but the course is not tailored towards them. If you want to work with images or networks, you might be interested in one of the following courses: “Computer Vision for Image and Video Data Analysis with Python” (15-19 September) or “Advanced Methods for Social Network Analysis” (15-19 September).

Learning objectives

By the end of the course, you will:

be able to identify research methods from computer science and computational linguistics which can be used for research in the domain of social science
have an understanding of the principles of supervised and unsupervised machine learning
be able to explain the principles of these methods and describe the value of these methods
know how to analyze textual data
have basic knowledge of the programming language Python and know how to use Python-modules for questions relevant in the domain of the social sciences
be able to independently analyze quantitative textual data using machine learning techniques

Prerequisites

Knowledge of basic statistics (linear and logistic regression)
Some experience with computational methods, programming in general, and/or statistical languages (but not necessarily Python) is highly recommended to participate in this course. During the first day of the course, we will discuss some fundamental aspects of coding in Python at a fast pace. In order to follow along, we recommend those who have little previous experience with computational methods or statistical languages to take part in the online blended learning course “Introduction to Computational Social Science with Python” (01-05 September).
Participants are expected to have a working Python environment installed (see below), and we strongly recommend that participants spend a couple of hours with one of the many free online resources to familiarize themselves with the very basics of Python to have an easier start. For a basic introduction or refresher to Python programming, participants may also consider taking the online workshop “Introduction to Python” that takes place from 25-28 August.

Software and Hardware Requirements

Participants should bring their own laptops for use in the course. You need to have a current Python environment installed and need to be able to install and update packages on your own. In the weeks prior to the course, the lecturers will distribute a setup guide which will ensure that participants can work within a similar environment to the lecturers. This guide is entirely optional, as you may alternatively set up your environment to your own preferences. All relatively recent versions of Python (in general, 3.8 or higher) should be fine. If you still have an older version, you may not be able to run the example code 1:1 but need to adapt it. Make sure you have recent versions of crucial packages such as pandas, numpy, scipy, scikit-learn, gensim, keras, sentence transformers and BERTopic installed. Additionally, it is advisable to have access to Google Colab and (Microsoft) GitHub. Therefore, please ensure that you have Google and Microsoft accounts, and can execute code through Google Colab.

Schedule

Day 1: Introduction and Getting Started in Python

Introduction and overview of the course
Principles of quantitative textual analysis for social scientists
Getting started with programming in Python: Introduction to the main concepts (such as data types, functions, and methods)
Practical discussion of benefits and drawbacks of working with different IDEs, as well as working with specific modules (such as pandas) versus native Python data structures
Conducting an exercise that focuses on setting up our first simple machine learning classifier

Suggested reading:

Kitchin, R. (2014). Big Data, new epistemologies and paradigm shifts. Big Data & Society, 1(1), 1-12. doi:10.1177/2053951714528481

van Atteveldt, W., Trilling, D., Arcila Calderon, C. (2022): Computational Analysis of Communication: A practical introduction to the analysis of texts, networks, and images with code examples in Python and R. Hoboken, NJ: Wiley. Chapters 1-4. https://cssbook.net/

Day 2: Preparing for Analysis: From Text to Features

Introduction to the toolkit accessible to social scientists working with 'big' textual datasets
Inductive and deductive approaches to computer-aided content analysis
Exploratory techniques to explore your data
When, why, and how do we pre-process?
Regular expressions
Natural Language Processing with NLTK and spacy
From text to features: count vectorizers and tf-idf vectorizers

Suggested reading:

Boumans, J. W. & Trilling, D. (2016). Taking stock of the toolkit. An overview of relevant automated content analysis approaches and techniques for digital journalism scholars. Digital Journalism, 1, 8-23. doi:10.1080/21670811.2015.1096598

Denny, M. J., & Spirling, A. (2018). Text preprocessing for unsupervised learning: Why it matters, when it misleads, and what to do about it. Political Analysis, 26(2), 168-189. doi:10.1017/pan.2017.44

Day 3: Unsupervised Machine Learning

Principles and techniques of Unsupervised Machine Learning techniques

e.g., a brief introduction to Principal Component Analysis, k-means clustering, and hierarchical clustering

Topic modeling with Latent Dirichlet Allocation (LDA)
Hands-on instructions to apply these techniques, using modules such as scikit-learn and gensim
Comparing techniques of unsupervised learning with supervised learning

Suggested reading:

Maier, D., Waldherr, A., Miltner, P., Wiedemann, G., Niekler, A., Keinert, A., ... Adam, S. (2018). Applying LDA topic modeling in communication research: Toward a valid and reliable methodology. Communication Methods and Measures, 12(2-3), 93-118. doi:10.1080/19312458.2018.1430754

Day 4: Supervised Machine Learning

Principles and techniques of Supervised Machine Learning
Discussion of how logistic regression and Naive Bayes classifiers can be used to predict, for instance, movie ratings or topics of news articles
Evaluation metrics (accuracy, precision, recall, ...)
Hands-on instructions to apply these techniques, using modules such as scikit-learn
Alternative models (e.g., Random Forests)
Advanced Supervised Machine Learning (e.g., cross-validation, grid search, model selection, and tuning)

Suggested reading:

Day 5: Recent Developments in Machine Learning

Visualization and presentation of findings
Outlook: Recent developments that are out of the scope of this course (e.g., embedding models, deep learning with keras, Transformer models, and more)

Suggested reading: