A study of 1.53 billion geotagged tweets finds that typical dictionary methods of assessing well-being using positively-connotated or negatively-connotated words produce results inconsistent with surveys of well-being and health in 1,208 counties in the United States; however, the removal of as few as three frequent, misleading words, such as "LOL," "love," or "good," can improve well-being predictions, according to the authors.
Article #19-06364: "Estimating geographic subjective well-being from Twitter: A comparison of dictionary and data-driven language methods," by Kokil Jaidka et al.
MEDIA CONTACT: Kokil Jaidka, National University of Singapore, SINGAPORE; e-mail: jaidka@nus.edu.sg ; Johannes Eichstaedt, Stanford University, CA; e-mail: johannes.stanford@gmail.com
###
Proceedings of the National Academy of Sciences