Evolving Consumer Responses to Social Issue Campaigns: A Data-Mining Case of COVID-19 Ads on YouTube

Abstract

Based on previous literature on comment-ranking algorithms and the role of popular opinion, we propose a data-mining approach to monitor evolving consumer responses to social issue campaigns. In particular, the proposed approach can (1) identify top-ranked comments on a social issue campaign in the dynamic social media environment and then (2) retrieve popular opinion from the top-ranked comments from a longitudinal perspective. To illustrate how to use the approach, we tracked the development of popular opinion contained in top-ranked comments posted about five COVID-19 brand videos that adopted different frames (i.e., employee appreciation, donation, call to action, frontline worker appreciation, and brand promotion). Results indicated that popular opinion resonates with the donation, frontline worker appreciation, and brand promotion frames, whereas popular opinion subverts the employee appreciation and call-to-action frames. The study has important methodological implications for advertising scholars and practitioners.

Keywords:

• Consumer responses
• data mining
• social issue ads
• social listening
• YouTube

Notes

1 The Python script for downloading YouTube comments is available at https://github.com/hridaydutta123/the-youtube-scraper. Note that the script might be outdated because YouTube updates the platform regularly.

2 In a pretest, using Selenium WebDriver, we downloaded 1,653 comments under a YouTube video in the same order as they were shown according to the “top comments” order on the video page at the data collection time. Then, we created a column titled “top order” to indicate their display order on the video page and a column titled “time order” to indicate their posting order. In multiple regression, we used the number of likes each comment receives and its “time order” to predict its “top order.” Results indicated that the more likes a comment has received (popularity rank) and the more recently it was posted (recency rank), the more likely the comment is ranked top. Moreover, the number of likes explains a much higher percentage (34%) of the variance in a comment’s “top order” than “time order” does (1.7%), shedding light on the extremely important role of likes in determining the top-ranking order of a comment.

3 The Python script of the proposed data mining approach is available here: https://github.com/ad-research-29/top-comments.

Additional information

Notes on contributors

Yang Feng

Yang Feng (PhD, Southern Illinois University Carbondale) is an associate professor of advertising, School of Journalism and Media Studies, College of Professional Studies and Fine Arts, San Diego State University.

Huan Chen

Huan Chen (PhD, University of Tennessee) is an associate professor of advertising, Department of Advertising, College of Journalism and Communications, University of Florida.