Elsevier

Poetics

Volume 41, Issue 6, December 2013, Pages 570-606
Poetics

Exploiting affinities between topic modeling and the sociological perspective on culture: Application to newspaper coverage of U.S. government arts funding

https://doi.org/10.1016/j.poetic.2013.08.004Get rights and content

Highlights

  • Demonstrates the usefulness of topic modeling for cultural sociology.

  • Uses LDA to assess coverage of 5 newspapers re: government arts funding (1986–1997).

  • 12 topics are identified and validated among nearly 8000 newspaper articles.

  • Demonstrates how such topic models can be used to gauge framing & nuances of meaning.

  • Also demonstrates how topic models reveal differences among sources of the articles.

Abstract

Topic modeling provides a valuable method for identifying the linguistic contexts that surround social institutions or policy domains. This article uses Latent Dirichlet Allocation (LDA) to analyze how one such policy domain, government assistance to artists and arts organizations, was framed in almost 8000 articles. These comprised all articles that referred to government support for the arts in the U.S. published in five U.S. newspapers between 1986 and 1997—a period during which such assistance, once noncontroversial, became a focus of contention. We illustrate the strengths of topic modeling as a means of analyzing large text corpora, discuss the proper choice of models and interpretation of model results, describe means of validating topic-model solutions, and demonstrate the use of topic models in combination with other statistical tools to estimate differences between newspapers in the prevalence of different frames. Throughout, we emphasize affinities between the topic-modeling approach and such central concepts in the study of culture as framing, polysemy, heteroglossia, and the relationality of meaning.

Introduction

This article addresses three puzzles. The first is methodological. How can researchers analyze large quantities of textual data efficiently and effectively? Specifically, how can we capture the information we need, reduce its complexity, and provide interpretations that are substantively plausible and statistically validated? We present topic modeling and, specifically, Latent Dirichlet Allocation (LDA) as a promising solution to these challenges.

The second puzzle is theoretical: The sociology of culture has long been theory-rich and methods poor. Sociologists who study culture have generated numerous theoretical insights and developed concepts that promise a deep understanding of cultural change. Yet they have often lacked the means to make such concepts operational (Mohr, 1998). We suggest that topic modeling may provide a way to do just that for such central concepts as framing, polysemy, heteroglossia, and the relationality of meaning.

The third is part of an ongoing study by the first author of the dramatic politicization of government support for arts organizations and artists in the late 1980s after a two-decade honeymoon. This article addresses how press coverage of public funding of the arts evolved from 1986 to 1997, a period that spans the beginning and end of the period of most acute contention.

Section snippets

Background: the crisis of public arts support in the U.S.

U.S. municipalities supported museums as early as the nineteenth century and financed bands and orchestras into the 1920s; the Roosevelt administration created a federal jobs program for artists during the Great Depression; several states established arts councils before 1965, and arts organizations receive substantial tax subsidy through the charitable deduction.1

Press accounts of public support for the arts in the U.S.

Did the tone of press coverage of government arts support change during the 1980s and 1990s? Did changes, if any, precede or coincide with political attacks on the NEA? Were some newspapers more likely to frame arts support in a positive light, whereas others were more likely to frame it as offensive or contentious? In short, how did the press respond to, participate in or contribute to the NEA's political woes?

Topic modeling: an inductive relational approach to the study of culture

Textual analysis has always been a central part of the study of culture. The digitization of huge quantities of text has raised the stakes by enabling scholars to launch more ambitious projects, while requiring development of new, more powerful, analytic tools. As a leading text on content analysis puts it, digitization shifts “the bottleneck of content analysis from the costs of access and tedious human coding to the need for good theory, sound methodology, and software…” (Krippendorf, 2004,

Topic modeling renders operational central ideas in the sociology of culture

Although automated approaches to textual analysis are increasingly plentiful, specific affinities between the topic-modeling algorithm and key ideas in the sociology of culture produce a strong fit between theory and method. In this section, we discuss the ways in which topic models enable scholars to render operational ideas about the relationality of meaning, heteroglossia, and framing.

Substantive application: did news outlets vary in their use of conflict frames?

For students of culture, producing a topic model is not an end in itself, but rather the first step in a longer process of interpretation and analysis. In the current case, by construing Topic 2 (arts controversies) and Topic 8 (polarization) as alternative frames for depicting conflict over arts grants—in one case treating controversial grants as a distinct problem, in the other treating arts funding as one of many related “social issues”—we can ask whether different news outlets varied in

Conclusions and further work

This article describes how to use probabilistic topic models of newspaper articles to study cultural trends, moods, and depictions. We studied press coverage of government grants supporting the arts between 1986 and 1997. During this period, such grants became controversial and the National Endowment for the Arts, the federal agency that made many of them, faced fierce attack from Republicans and conservative social movement organizations. In this section we first summarize key substantive

Acknowledgements

Research support from Princeton University and sabbatical support for the first author from the Russell Sage Foundation are gratefully acknowledged, as is the assistance of Brian Steensland in gathering parts of the textual data. Support from the Rockefeller Foundation and the Andrew W. Mellon Foundation (through a grant to Princeton's Center for Arts and Cultural Policy Studies) for data collection, research and sabbatical support from Princeton University and the Russell Sage Foundation and

Paul DiMaggio is A. Barton Hepburn Professor of Sociology and Public Affairs at Princeton University. His interests include patterns of cultural participation, the organization of the Internet, formal and statistical methods of cultural analysis, and the impact of social networks on social inequality.

References (72)

  • G.E.P. Box

    Robustness in the strategy of scientific model building

  • J. Alexander

    Command Performance: An Actress in the Theatre of Politics

    (2000)
  • M.M. Bakhtin
  • R.D. Benford

    An insider's critique of the social movement framing perspective

    Sociological Inquiry

    (1997)
  • R. Benson

    Shaping Immigration News: A French-American Comparison

    (2013)
  • S.L. Bird

    Seeking the audience for news: response, news talk, and everyday practices

  • D.M. Blei

    Probabilistic topic models

    Communications of the ACM

    (2012)
  • D.M. Blei et al.

    Topic models

  • D.M. Blei et al.

    Supervised topic models

    (2007)
  • D.M. Blei et al.

    Latent dirichlet allocation

    Journal of Machine Learning Research

    (2003)
  • P.J. Boczkowski

    News at Work: Imitation in an Age of Information Abundance

    (2010)
  • M. Brenson

    Washington's stake in the arts

    (1998)
  • M.L. Buis

    FMLOGIT: Stata Module Fitting a Fractional Multinomial Logit Model by Quasi Maximum Likelihood

    (2008)
  • C. Burgess

    Multiple streams and policy community dynamics: the 1990 NEA Independent Commission

    Journal of Arts Management, Law, and Society

    (2006)
  • A. Copestake et al.

    Semi-productive polysemy and sense extension

    Journal of Semantics

    (1995)
  • P. DiMaggio

    Introduction

  • P. DiMaggio

    Culture and cognition

    Annual Review of Sociology

    (1997)
  • P. DiMaggio

    The myth of culture war: the disparity between private opinion and public politics

  • P. DiMaggio et al.

    Public attitudes toward cultural authority and cultural diversity in higher education and the arts

  • P. DiMaggio et al.

    The role of religion in public conflicts over the arts in the Philadelphia Area, 1965-1997

  • P. DiMaggio et al.

    Public Opinion and Political Vulnerability: Why Has the National Endowment for the Arts Been Such an Attractive Target?

    (1999)
  • Doyle, G., Elkan, C., n.d. Financial Topic Models. http://www.umiacs.umd.edu/∼jbg/nips_tm_workshop/22.pdf (accessed...
  • G. Doyle et al.

    Accounting for burstiness in topic models

  • S.C. Dubin

    Arresting Images: Impolitic Art and Uncivil Actions

    (1992)
  • S. Feldman

    Enforcing social conformity: a theory of authoritarianism

    Political Psychology

    (2003)
  • O. Fiss

    State activism and state censorship

    Yale Law Journal

    (1991)
  • J. Frohnmayer

    Leaving Town Alive: Confessions of an Arts Warrior

    (1993)
  • W.A. Gamson

    Talking Politics

    (1992)
  • W.A. Gamson et al.

    Media images and the social construction of reality

    Annual Review of Sociology

    (1992)
  • A. Gelman et al.

    Posterior predictive assessment of model fitness via realized discrepancies

    Statistica Sinica

    (1996)
  • General Social Survey

    Retrieved July 14, 2013 from the iPOLL Databank

    (1998)
  • A. Gelman et al.

    Bayesian Data Analysis

    (2003)
  • M. Gentzko et al.

    What Drives Media Slant? Evidence from U.S. daily Newspapers. NBER Working Paper 12707

    (2007)
  • S.M. Gerrish et al.

    A language-based approach to measuring scholarly impact

  • S.M. Gerrish et al.

    Predicting legislative roll call votes from text

  • T. Griffiths et al.

    Finding scientific topics

    Proceedings of the National Academy of Sciences

    (2004)
  • Cited by (649)

    View all citing articles on Scopus

    Paul DiMaggio is A. Barton Hepburn Professor of Sociology and Public Affairs at Princeton University. His interests include patterns of cultural participation, the organization of the Internet, formal and statistical methods of cultural analysis, and the impact of social networks on social inequality.

    Manish Nag is a PhD candidate in the Department of Sociology at Princeton University. His research utilizes computational social science innovations in text analysis and social network analysis to understand cultural change in media and academic discourses, as well as change and resilience in global networks of people, goods and ideas.

    David Blei is an associate professor of Computer Science at Princeton University. His research focuses on probabilistic topic models, Bayesian nonparametric methods, and approximate posterior inference. He works on a variety of applications, including text, images, music, social networks, and scientific data.

    View full text