MIT Algorithm Predicts What Makes Photos Popular On the Web
Source: BARRIBEAU

By analyzing millions of images from Flickr, an MIT researcher has discovered some of what makes an image popular―and frankly, it's more or less what we were expecting.

Aditya Khosla, a PhD candidate at MIT's Computer Science and Artificial Intelligence Lab analyzed 2.3 million images from Flickr, and through a combination of both automatic image analyses, image content, and social cues, was able to predict how well each image would do. And while the data comes from Flickr, it probably holds just as true for most other image based social networks, too.

There's a simplified version of the calculator available online (though with all the attention, it's been slightly overwhelmed). For a more in-depth look at the analysis and predictions, you should dive into the paper itself. Here are some factors that Khosla identified:Colors:

We observe that on average, the greenish and bluish colors tend to have lower importance as compared to more reddish colors.

Content:

• Strong positive impact: miniskirt, maillot, bikini, cup, brassiere, perfume, revolver

• Medium positive impact: cheetah, giant panda, basketball, llama, plow, ladybug

• Low positive impact: wild boar, solar dish, horse cart, guacamole, catamaran

• Negative impact: spatula, plunger, laptop, golfcart, space heater

Social:

•Mean Views: mean of number of normalized views of all public images of the given user

• Photo count: number of public images uploaded by the given user

• Contacts: number of contacts of the given user

• Groups: number of groups the given user belongs to

• Group members: average number of members in the groups a given user belongs to

• Member duration: the amount of time since the given user joined Flickr

• Is pro: whether the given user has a Pro Flickr account or not

Anecdotally, this matches up with a lot of what we're used to seeing. The popularity of items related to scantily clad women and cute animals; bright and poppy colors; and already being a well-established user inside of Flickr's community. One thing that would be interesting is to see if the same factors translate across other social networks universally. 500px tends to have a different aesthetic than Flickr to Instagram to VSCO Cam to Facebook. Do these truths transfer, or are they Flickr specific?



Meanwhile, the online algorithm gave this shot of my cat a popularity of 4.146, which, quite frankly, I'll take.