Machine Learning for Social Scientists

This class teaches supervised machine learning methods, i.e. statistical methods for prediction and description tasks. We will focus on methods that are relevant to social science, especially ones using text as data. The programming language used is R.

We will cover (1) text-as-data skills such as preprocessing, TF-IDF, dictionary methods and keyword selection; (2) basics of supervised learning such as performance metrics, train-test splits, cross-validation and regularization; (3) prediction methods such as logistic regression, softmax regression, Naive Bayes models and support vector machines, plus scaling methods like item-response theory models; (4) neural networks and their architecture, including deep neural nets, word embedding models and transformers; and finally (5) applications of neural nets to text-as-data (large language models, text classification).