The way we see ourselves — cheerful, cool, daring — often translates into the products we buy, which in turn creates a “personality” for the brands we consume.
“Brand personality scales” have been around for many years, using consumers’ feedback to attribute human characteristics to companies. These scales, which find that Cracker Barrel is “wholesome” and Sephora is “contemporary,” have proven to be reliable marketing tools.
Now, a team including an Arizona State University professor and IBM researchers have harnessed machine learning to accurately predict brand personality ratings by analyzing hundreds of thousands of social media posts.
Yili Hong, an associate professor of information systems in the W. P. Carey School of Business at ASU, has done a lot of big data research on social media and crowd funding. His latest paper was published in the Journal of Management Information Systems.
“People self-select to buy brands because they identify with certain characteristics of those brands,” Hong said. “And the brands reinforce that by communicating those characteristics to the public through their advertisements.”
Typically, these characteristics are measured through a survey of consumers, which is expensive and provides a one-time snapshot.
Hong and his team massively scaled that up by mining social media posts to determine consumers’ views, and they discovered their model was able to accurately predict companies’ “brand personalities.”
For a few years, data researchers have been analyzing social media text content to determine human characteristics, but Hong’s team is the first to apply the model to brand personalities. To do that, they examined three sources of data: Twitter profiles and two months’ worth of tweets from nearly 2 million brand followers, 680,000 tweets from the companies themselves and 312,000 employees’ reviews on Glassdoor.com.
The data set was huge and the IBMThe study found IBM’s personality to be “intelligent,” “technical” and “corporate.” computer analyzed millions of words to come up with the “personalities” of 219 brands.
“This was a very large endeavor,” Hong said. “We were able to use state-of-the-art machinery.”
The team also did a typical online survey of more than 3,000 consumers, who ranked the 219 brands for various personality attributes. Then the researchers compared that result to the social media analysis and discovered that the computer model was up to 79% accurate for some attributes.
Among the results: Disney is “family-oriented” and “friendly,” Apple is “cool” and “independent,” ESPN is “masculine,” Tiffany is “glamorous” and the North Face is “rugged.”
So will social media text analysis replace the old-fashioned survey? Hong thinks not.
“There will always be a place for the survey,” he said.
“You need to have an existing truth so you can build the model and we need surveys to get the truth. You need the human collaborator to gather these ground truths so you can apply the model and scale it to other scenarios.”
A company might do a consumer survey to gauge its brand personality. But the firm also wants to know how that personality evolves over time, how it compares to a competitor and whether it changes after a new advertising campaign. So Hong’s team also created a prototype cloud-based system that companies could use to collect social media data and analyze it to monitor their own brand personality on an ongoing basis.
“It’s very costly to do surveys over and over again, but based on this technique, we can get the information,” Hong said.
One of the study’s authorsBesides Hong and Hu, who is now an assistant professor at the University of Illinois at Chicago, the other authors are David Gal of the University of Illinois at Chicago and Anbang Xu, Vibha Sinha and Rama Akkiraju, all of IBM. is Yuheng Hu, who earned his PhD in computer science at ASU before going to work at IBM. He and Hong collaborated on previous big data research, and Hong believes the cross-disciplinary approach is important.
“There is a lot that computer science can offer to business because a lot of abstract concepts nowadays can be measured using computational modeling,” he said.
“It's very popular these days to talk about human and machine collaboration because the algorithms cannot run by themselves. This is use-inspired work to solve problems.”
Top image by Pixabay
More Science and technology
More than 60 distinct viruses found in feces of common park duck
Billions or even trillions of tiny microbes, like bacteria, fungi and viruses, live inside every single animal, making each one of us hosts to entire ecosystems. Those invisible microbes outnumber…
ASU researchers first to fully sequence Gila monster genome, thanks to crowd-funding campaign
The Sonoran Desert is full of wild creatures, from sharp-tailed scorpions that glow under black light to desert toads that secrete hallucinogenic toxins from their skin. Perhaps no creature, though,…
Sparking an evolution in robotics
Thinking about swarms of robots might conjure up images from old sci-fi movies in which Earth is invaded by armies of mechanical androids from another world that are programmed and weaponized to…