Developing and Validating the Moral Foundations Dictionary for News Narratives: A Crowd-Sourced Approach

• In Event: New Directions in Text Mining

Abstract

An increasing number of studies has been utilizing the Moral Foundations Dictionary (MFD; Graham, Haidt, & Nosek, 2009) to extract moral information from textual data. Yet, the MFD inherits certain limitations such as ad hoc pre-selection and overlappings of word stems that limit its validity. In this paper, we introduce a crowd-sourced approach for developing a moral foundation dictionary derived from a total of 65,012 highlights of 1,010 online newspaper articles highlighted by a U.S. representative sample of 557 human coders. We utilized various methodologies and parameters for creating four diverse moral foundations dictionaries and highlight approaches for assessing their convergent and predictive validity by correlating extracted word counts with other computational content measures such as themes and sentiment, as well as real-world behavioral outcomes such as sharing counts of newspaper articles. We discuss further refinements and tunings of our dictionaries and point towards future research directions for their implementation.

Authors

• Frederic Rene Hopp, U of California, Santa Barbara

• James Michael Mangus, U of California, Santa Barbara

• Reid Swanson, U of Southern California - Institute for Creative Technologies

• Andrew Gordon, U of Southern California - Institute for Creative Technologies

• Peter Khooshabeh, U of California, Santa Barbara

• Rene Weber, U of California - Santa Barbara