Advances in Deep Learning Methods for Text, Image, and Video Data

Session Submission Type: Full Paper Panel

Session Description

Deep learning tools hold great promise for political science research. This panel, one of two linked panels using deep learning for the analysis of big data, focuses on advances in machine learning methods. The authors tackle important questions on how to efficiently and accurately process image, text, and video data. Joo and Steinert-Threlkeld provide an overview of supervised deep learning for visual content analysis. Zhang and Peng also focus on image data, emphasizing the potential benefits of unsupervised image clustering. Dietrich, Ko, and Sen discuss a new technique for working with video data, with an application to 2015 United Kingdom election. Anastasopoulos contributes a helpful corrective and critique of the use of deep learning for text analysis, describing both the benefits and costs of these techniques. In sum, the panel provides an introduction to new methods while also pushing the frontier of these tools.

Sub Unit

• Division 8: Political Methodology

Individual Presentations

• Image as Data: Automated Visual Content Analysis for Political Science - Jungseock Joo, UCLA; Zachary Steinert-Threlkeld, University of California - Los Angeles

• Traffic Cameras, Voting Costs, and the 2015 United Kingdom General Election - Bryce Dietrich, University of Iowa; Hyein Ko, University of Iowa; Payel Payel Sen Sen, Stony Brook University

• When You Should(n’t) Use Deep Learning with Text - Jason Anastasopoulos, University of Georgia

• Unsupervised Clustering of Image Data: Overview and Assessment of Performance - Han Zhang, The Hong Kong University of Science and Technology; Yilang Peng, University of Georgia

Chair

• Jennifer Pan Stanford University

Discussants

• Michelle Torres Rice University

• Andreu Casas Vrije Universiteit Amsterdam