Fearless Idea 12: Filter Social Media Bias

SMITH BRAIN TRUST — Social media users post millions of “likes” and comments every year on brand pages for everything from AAA to Zyrtec. That’s a potentially rich source of information for marketers trying to gauge customer sentiment, but built-in biases create challenges. New research from professor Wendy W. Moe and a colleague at the University of Maryland’s Robert H. Smith School of Business shows how to scrub the data and produce reliable scores.

Moe and her co-author, Smith School professor Kunpeng Zhang, base their brand favorability measure on an analysis of 7 billion Facebook “likes” and comments from 170 million users across more than 3,000 company pages. They then compare results to traditional brand tracking surveys and find a strong correlation.

“This provides validation that our brand favorability score is an accurate and unbiased measure of the brand,” the authors write in a working paper. The key to filtering out social media noise is to account for biases related to user traits, community size and brand type.

Most users who express opinions in online communities tend to lean negative or positive, which can skew results in one direction or the other. To adjust for potential bias, Moe and Zhang look at user behavior across brands. Negative comments mean less when coming from unhappy customers who frequently trash other brands, while positive comments mean less when coming from habitual cheerleaders.

Community size also matters. Smaller, more homogenous communities tend to reinforce positive opinions, creating an echo chamber effect. “The smaller your community, the more positive the bias,” Moe says. “And the larger your community, the more negative the bias.”

Marketers must also account for brand type. Commercial brands tend to attract more negativity than other types. “One explanation is that people who follow celebrities, sports teams, nonprofit organizations and other special interests may be ‘super fans’ and unlikely to provide negative feedback,” Moe says.

Read more: “Bias on Your Social Media Brand Page: How Do We Measure Underlying Brand Favorability and Identify Potential Bias in Sentiment Metrics?” is a working paper.

Wendy W. Moe is a professor of marketing and director of the Smith School’s Master of Science in Marketing Analytics

Research interests: Models for online and social media marketing; consults for numerous corporations and government agencies, helping them develop and implement state-of-the-art statistical models in the area of web analytics, social media intelligence and forecasting; frequently serves as an expert witness in litigation.

Selected accomplishments: Research in web analytics was the foundation for NetConversions, an early innovator in the area of online data collection and analysis. She was part of the founding team that brought the company from startup to acquisition in 2004. Author of Social Media Intelligence (Cambridge, 2014). Serves on the Board of Trustees for the Marketing Science Institute and the advisory board for the Wharton Customer Analytics Initiative.

About this series: The Smith School faculty is celebrating Women’s History Month 2017 in partnership with ADVANCE, an initiative to transform the University of Maryland by investing in a culture of inclusive excellence. Daily faculty spotlights support activities from the school’s Office of Diversity Initiatives, culminating with the sixth annual Women Leading Women forum on March 30, 2017.

Other fearless ideas from: Rajshree Agarwal  |  Ritu Agarwal  |  Leigh Anenson  |  Kathryn M. Bartol  |  Christine Beckman  |  Margrét Bjarnadóttir  |  M. Cecilia Bustamante  |  Rellie Derfler-Rozin  |  Waverly Ding  |  Wedad J. Elmaghraby  |  Rosellina Ferraro  |  Rebecca Hann  |  Amna Kirmani  |  Hanna Lee  |  Hui Liao  |  Wendy W. Moe  |  Courtney Paulson  |  Louiqa Raschid  |  Rebecca Ratner  |  Rachelle Sampson  |  Debra L. Shapiro  |  Cynthia Kay Stevens  |  M. Susan Taylor  |  Vijaya Venkataramani  |  Janet Wagner  |  Yajin Wang  |  Yajun Wang  |  Liu Yang  |  Jie Zhang  |  Lingling Zhang  |  PhD Candidates