How guns and bikinis make your online photos more popular
|
Version 0 of 1. Next time you're tempted to share a photo of a golf cart or a spatula, don't. New research has confirmed those nagging suspicions that your followers won't be interested. Computer scientists analysed the colour, texture, content and context of more than 2m Flickr photos, then looked at the number of views, to figure out what factors determine a picture's chances of internet fame. Most helpfully, the research contains lists of objects that positively or negatively affect a photo's appeal. Objects with a "high positive impact" included revolvers, miniskirts, bikinis, bras, perfume and cups (so full speed ahead with your latte art Instagram). Meanwhile, spatulas, plungers, laptops and golf carts were turnoffs. The objects with a "medium positive impact" were more surprising: apparently, llamas, cheetahs, ladybirds and basketballs will get you clicks, as will images in "reddish" colours. Horse carts, guacamole, wild boar and solar dishes have a slight positive impact, but won't win as many views as guns or lingerie. The authors of the research paper, What Makes an Image Popular?, which was presented at April's World Wide Web conference, claim it is the first attempt to analyse both the content and context (title, online tags and user information) of an image to predict its reach. Using the same methodology the team have developed an online tool to estimate photos' likely popularity before they are posted online. An algorithm scores submitted photos out of 10, and uses this score to predict how many views it will receive on the image-sharing site Flickr (a score of four corresponds to 16 views per day). MIT PhD student Aditya Khosla, one of the paper's authors, advises comparing the scores of several images before posting to avoid embarrassing social media flops. The researchers believe their findings could be used to develop a "photography popularity tool" that would analyse colour, object and texture information to provide photographers with "suggestions on how to modify their pictures for broad appeal". For now though, the current algorithm has its limits – it recognises only 1,000 objects, and doesn't bump up images of celebrities or memes. This may explain the low results of some undeniably popular viral images: Doge scores a measly 4.196 out of 10, while the Ellen DeGeneres Oscars selfie (Twitter's most-shared photo of all time) gets a disappointing 4.158. Just think how popular it could have been if one of them had been holding a revolver. |