MIT algorithm predicts photo popularity
|We upload this image to Khosla's popularity calculator and got a score of 4.176.|
Ask the average social media user what makes a photo popular online and they'll probably answer 'cats' or 'pretty women'. That gut feeling has now been confirmed by MIT PhD candidate Aditya Khosla who has analyzed more than 2.3 million images hosted on Flickr to answer the question 'what makes an image popular'?
Khosla used a mix of automation and observation to determine something most people already suspected: unique and/or sexy images do well, dull images do not. Bikinis, for example, rate above golf carts.
Naturally this is an oversimplification, the original paper on his research is a good bit more detailed and fascinating to anyone marketing images in a social media environment. One of the paper's key sections indicates color preferences as they relate to popularity, something that could help a photographer or marketing team craft images with more reach.
He also released a calculator to evaluate any image and score it for its potential for popularity, though the site is being bogged down with traffic as of this writing.
His data came from Flickr's vast archive of images, so it's not clear if the same patterns would hold true for different photo sharing platforms. Still, it's an interesting scholarly look at that question we've all thought when browsing the images on social media 'why did people 'like' that'?