Deep Learning and Ideological Rhetoric Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 19, 2019
  • Acree, Brice
    • Affiliation: College of Arts and Sciences, Department of Political Science
  • Political conflict unfolds in language. To understand the quest for, and exercise of, power, we must understand political speech. For more than a decade, political methodologists have sought to develop methods for using language to study how politicians speak to the public, to the media, and to each other. This dissertation advances that line of research. Chapters 2 and 3 address `deep learning' methods for analyzing political texts. These computational models, built on artificial neural network architectures, seek to automatically learn complex patterns in data. Chapter 2 introduces basic deep learning models, and Chapter 3 introduces distributional word representations and convolutional neural network models. The final chapter points these methods toward a substantive problem. As a discipline, we have studied, and debated, the nature of political ideology. We have often found that, for citizens and elites alike, ideology is oriented around a single primary dimension: liberalism to conservatism. In Chapter 4, I offer a different view. Using the language from political ideologues and presidential candidates, I show that ideology can be much richer and varied than a single dimension will capture, but that political debate compresses ideological expression into a single dimension.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Carsey, Thomas M.
  • Stimson, James
  • MacKuen, Michael
  • Ryan, Timothy
  • Gross, Justin
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2016

This work has no parents.