Selfies are so hot that they might already be over (if I’m aware of something being cool it usually isn’t anymore) but you don’t have to just crank out pictures of your mug taken at arm’s length indiscriminately in hopes of finding an angle that plays well on social media. Thanks to MIT PhD student Aditya Khosla (via CSAIL), you’ll save yourself some arm strain with shortcuts to selfie virality.
Using data culled from the analysis of 2.3 million Flickr photos, Khosla developed an algorithm that can predict the popularity of a photo (as it its actual view count) with a high degree of accuracy, based not just on a user’s amount of social influence (previous models already do that) but also factoring in image content. Khosla’s algorithm takes into account factors like color, types of objects included, tags and more to work out how many people will like, look at or share that image.
The news of what works isn’t particularly earth-shattering, based on Khosla’s early recommendations: sexy shots work best, unsurprisingly, at least in terms of driving views. Those that feature prominent displays of “bikinis, bras and mini-skirts” did the best. Lucky for selfie-shooters, those photos with people in them also did much better than those without.
The ultimate aim isn’t just to provide tips about what photos do best on the web – Khosla hopes to eventually be able to automatically modify pictures with edits that make them more popular. If it works, that would put a whole new spin on mining Instagram for likes. Vanity, you have a new champion.