Meet the Algorithm That Can Predict Your Photo's Popularity Before You Post It
Social media can be a funny place. Why does one post spread like wildfire, while another languishes in obscurity?
The science behind a post's popularity has understandably gotten a lot of attention of late – for businesses investing in social media advertising, it's important to know that statistically, a post that includes an image uploaded on a Friday will provide the highest engagement rate. (Want more information on the best methods and times to post on various platforms? Check out this infographic).
Images, of course, aren't created equal. A recent study by researchers at Georgia Tech analyzed 1.1 million randomly selected Instagram photos and found that those with human faces had a 38 percent greater chance of being 'liked,' and were 32 percent more likely to get comments.
A new study out of MIT gets even more granular. Researchers, led by Aditya Khosla, analyzed 2.3 million Flickr photos to develop an algorithm that can reliably predict how many times a photo will be viewed based on social context (how many followers a user has) and the image's content.
The algorithm factors in variables including color (warm, bright shades like yellow and pink draw more views than cool, soft tones) and the type of object featured.
Unsurprisingly, sexy images attract the most eyeballs. (Miniskirts, bikinis, bras, and revolvers significantly increased a photo's popularity, the study found. Plungers, laptops and golfcarts, on the other hand? Not so much.)
While an image's content matters, social context still carries more weight when predicting popularity. "Overall, popularity is a difficult metric to understand precisely based on image content alone because social cues have a large influence on the popularity of images," the authors wrote. In other words, a picture of a plunger posted by a user with thousands of followers will probably still get more views than an image of a bikini-clad girl posted by a user with five.
Related: The 7 Secrets to Shareable Content
You can analyze your unpublished photos using a prototype on Khosla's research page, which measures an image's potential popularity on a relative scale from one to 10. It's a relatively crude measurement, but Khosla and his team plan to refine the algorithm.
"From an application standpoint, is there a photography popularity tool that could be built here?" they wrote. "Can photographers be aided with suggestions on how to modify their pictures for broad appeal vs artistic appeal?"
Khosla has already developed a similar product. Last year, he was part of the research team that created an algorithm able to automatically modify actors' headshots in order to make the photos more memorable.