Computational Social Science: New Approaches

Session Submission Type: Full Paper Panel

Session Description

In recent years, the increased availability of data and the advancement of statistical methodology have created an unprecedented opportunity to address longstanding questions in political science. This panel brings together four projects that exhibit new advances in methods and applications of computational social science, and demonstrate how we can analyze unstructured data, such as texts and videos, in ways that are useful for understanding political phenomena.

Eshima, Imai, and Sasaki empirically demonstrate that providing topic models with keywords can dramatically improve their performance. In the applications, they show that the proposed method provides more interpretable results, has better document classification performance, and is more robust to the number of topics than the standard topic models.

Knox et al. study the effects of online video using a controlled design that randomly manipulates the algorithm used to recommend videos to subjects. By experimentally injecting diversity into the stream of naturalistic suggestions on their own online video platform, the experiments simultaneously consider interacting and self-reinforcing processes in the supply, demand, and persuasive effects of media consumption.

Roberts and Yang examine how censorship has affected the development of Wikipedia corpuses, which are regularly used as training data that provide inputs to NLP algorithms. They show that word embeddings trained on the regularly censored Chinese language Wikipedia have very different associations between adjectives and a range of sensitive concepts, and people and historical events in China than its uncensored counterpart Baidu Baike.

Rodriguez, Spirling, and Stewart introduce a new approach based on word embeddings that allows researchers to make statements about how a word's usage and meaning varies over contexts such as time, partisan identity, or other document-level covariates. Their approach also enables researchers to conduct statistical inference and to make statements about uncertainty.

Sub Unit

• Division 8: Political Methodology

Individual Presentations

• Embedding Regression - Pedro L. Rodriguez, Vanderbilt University; Arthur Spirling, New York University; Brandon Michael Stewart, Princeton University

• Censorship’s Implications for Artificial Intelligence - Margaret E Roberts, University of California, San Diego; Eddie Yang, University of California, San Diego

• The Effects of Online Video Personalization Algorithms - Dean Knox, University of Pennsylvania; Andy Guess, Princeton University; Brandon Michael Stewart, Princeton University; Jason Anastasopoulos, University of Georgia; Adam J. Berinsky; Matthew A. Baum, Harvard University; Justin de Benedictis-Kessner, Harvard University; Christopher Lucas, Washington University in St. Louis; Naijia Liu, Princeton University

• Keyword Assisted Topic Models - Shusei Eshima, Harvard University; Kosuke Imai, Harvard University; Tomoya Sasaki, MIT

Chair

• Yuki Shiraito University of Michigan, Ann Arbor

Discussant

• Michelle Torres Rice University