Deep Learning in Political Science

Session Submission Type: Full Paper Panel

Session Description

Deep learning has shown us how to perform tasks with computers previously thought possible only with humans: naturally enough, there is much excitement about such artificial neural network
models in political science. With this in mind, this panel draws together four papers at the cutting edge of the application of these approaches to demonstrate best practice in political methodology. Collectively, we look "beyond the hype" with clear-eyed assessment of what works and what doesn't in this burgeoning literature. The papers inspect performance for both supervised and unsupervised tasks, and in multiple political contexts and settings, including social media in China and the US, experimental treatments and institutional speeches in the UK, Spain and France. In the first paper, Roberts and Sanford consider whether how well we can model the way that humans make real decisions on political questions, by using recurrent convolutional NNs. Then, Pineda and Mebane show how we can combine text and images into a single supervised learner for prediction in US politics using a multi-layer perceptron. Pan and Zhang attack a similar problem for Chinese social media, where censorship is common. Finally, Rodriguez and Spirling provide a comprehensive practical guide to the use of word embeddings, a
popular deep learning representation of texts.

Sub Unit

• DIVISION 8: Political Methodology

Individual Presentations

• Mining Human Decision Making - Margaret E. Roberts, University of California, San Diego; Luke Sanford, UC San Diego

• Using Neural Networks to Classify Based on Combined Text and Image Content - Walter R. Mebane, University of Michigan, Ann Arbor; Alejandro Pineda, University of Michigan

• CASM: A Deep-Learning Approach for Identifying Collective Action Events - Jennifer Pan, Stanford University; Han Zhang, Princeton University

• Word Embeddings: Properties, Performance and Prospects - Pedro L. Rodriguez, Vanderbilt University; Arthur Spirling, New York University

Chair

• Kevin M. Quinn University of Michigan, Ann Arbor

Discussant

• Jacob M. Montgomery Washington University in St. Louis