As in control as we like to think we are when we swipe through our Instagram feed and double-tap a photo, the reality is we’re a lot more habitual with our social feedback than we realize. So says a new study out of the University of Pennsylvania's Annenberg School for Communication, published in the International Journal of Communication.
Yilang Peng, a doctoral candidate at Annenberg, pitched a research project to Professor of Communication and Psychiatry John Jemmott approximately two years ago, suggesting they use computer vision—a computer science tool not often applied to visual communications research, says Peng—to unfurl the tightly-wound mystery of what makes food photos popular or go viral. The goal was to explore visual culture, which he says is increasingly ubiquitous in today’s media environment, by examining photo elements like color, composition, and the combination of taste and health perception.
Peng and Jemmott recruited 401 participants from a crowdsourcing platform to engage with a manufactured news feed featuring a randomized assortment of 15 food photos and 20 nonfood photos; food photos were selected to be diverse, according to the analysis of the computer vision. They measured “popularity” by evaluating “likes” and comments, while evaluating “viral” by looking at how likely respondents said they’d be to share a photo.
The results, in a nutshell: Photos that adhere to the “rule of thirds,” a common photography composition principle, were likely to be more aesthetically appealing. A moderate amount of color variety increased likability and aesthetic appeal, but color intensity did not. Arousing colors, like red and orange, made a photo more likable. Photos more visually complex—think: a bowl of different fruits—generated greater share intention and likability, and, surprisingly, both healthy foods and unhealthy foods tended to be likeable.
With the latter, the “why?” can be explained by psychology.
“Previous research has shown arousal is very important in driving viral behaviors, such as ‘liking,’ so it’s not surprising that tasty and unhealthy foods are more arousing,” Peng says. “But, people also want to maintain a positive self-image, and so liking healthy food is one way of doing that. People tend to like things on social media to express their values, identities, or desired selves.
“Healthy food is a way to make people feel good about themselves,” he adds.
The results of the study are significant in that they might inform ways to make healthy eating more popular. At present, Peng says, unhealthy foods, like desserts, tend to clog Instagram, potentially presenting misleading dietary social norms. Social media users get stuck in what Peng calls a “feedback loop,” in which the tasty-but-unhealthy photo attracts likes and comments, so they continue to post similar photos.
Findings of the study also reinforce the notion of color interpretation based on context.
“If you have a food photo that’s blue, people won’t like it—it’s calm [and not arousing], but also, food psychology shows the reason we like a color like red, associated with food, is because fruits turn red when they become ripe,” Peng explains. “We’ve evolved to develop this vision that favors red over green or blue in food.
“Red might be a [positively perceived] color in food, but blood, it might not be,” he says.
Furthermore, Peng adds, the study is a remarkable example of computer vision’s effectiveness for performing visual analysis. In this case, it was used to efficiently and accurately measure the colorfulness of a photo—say, how much red is included—and composition. That allows for more reliable data than subjective evaluations of how much color might be in a photo.
Building upon the food study, Peng is next using computer vision to explore the Instagram accounts of politicians. Using facial recognition, he’ll determine which photos tend to be more popular, based on visual cues like smiles and crowds in photos. That research is ongoing.