Computational approach to studying media coverage of organizations

  Media coverage of organizations is a social science topic that attracts attention from various domains. While there is a clear opportunity to expand this research using computational approach, the field lacks practical agreements on how to reconcile methodological tensions between norms and standards of social science and computational science. The purpose of this article is to start establishing methodological standards for social science research using large media coverage data across multiple organizations. We pose three key questions: When is the computational approach effective in assessing media coverage of organizations? What are practical challenges that social scientists face while using the computational approach to studying the media coverage of organizations? How effectively does a computational approach perform to measure the major media variables of volume and sentiment? We start by suggesting that this approach is particularly useful when the media coverage of organizations either entails context-dependence or involves a rare phenomenon. We then focus on demonstrating a replicable computational methodology. We detail data collection and analysis for studying the media coverage of 10,749 venture-capital-backed startup companies in the U.S. from 1980 to 2018, using 745,216 unique articles. The preliminary findings suggest that the computational analysis of media volume is highly valid and provides valuable insights for social scientists. In contrast, using popular dictionaries to analyze media sentiment across multiple organizations is problematic, suggesting the need to develop new methodologies. We conclude that social scientists and computer scientists should proactively collaborate to create tools to advance future research in this domain.