Get Your Social Media Pictures Down to a Science (Well, an Algorithm)
One day, because a writer can work only for so long on content for the sparkly new QLP website before needing a break, I sat down and looked at pictures. Actually, I did a little more than that (just a little): I sat and reviewed the list of users that Quality Logo Products was following on Instagram.
It turns out that during my months of managing the account, Quality Logo Products had become a fan of pugs.
Our company wasn’t the only group of pug fans, though. Not by far. The user known simply as “pugs” had 646, 961 followers at the time of this writing. Their most recent photo as I’m writing this, of a pug puppy glancing desperately at a treat, had about 62,100 likes.
When it comes to posting images to social media and getting likes, some truths are as eternal as anything gets online. Tiny, adorable animals get a lot of love, for example. Such truths are easy to come by through simple observation. Go check your Facebook feed, and let me know what you see.
What if there was a way to quantify those observations, though? What if your pictures could receive a score that would tell you exactly how popular they were likely to be?
That’s exactly what Aditya Khosla, a Ph.D. candidate at MIT and Facebook Fellow (yes, Facebook awards Facebook Fellowships) has set out to provide.
Popularity Ain’t Nothin’ but a Number?
In a paper released last month, “What Makes an Image Popular?” Khosla and his research team announced that they had devised an algorithm that would assign an image a score. That score would suggest how many views the photo was likely to get on Flickr.
It’s important to note that Khosla’s work focused on Flickr. The results of his work have yet to be scientifically tested on other social networks. That being said, the principles he confirmed make sense when considering networks like Instagram.
What Khosla confirmed is that image popularity depends on a number of factors, because of course things are never simple. Khosla, however, was able to divide the factors into two categories:
- Image content, the stuff actually captured in the picture, like cats or pancakes, along with the the subject’s surroundings.
- Social cues, the details like how many followers a user has and how many photos that user already has posted.
Not only did Khosla divide influences into those two categories; he also demonstrated numerically how important each category was. Additionally, his research took into account three possible ways a viewer could find images:
- An image comes up among the search results for a certain term or hashtag.
- Images from a viewer’s contacts or “friends” show up in the viewer’s news feed.
- A viewer browses through another user’s photo albums or galleries (in which case the popularity score compares only images contained within one user’s gallery).
Guess what Khosla determined? Social cues are incredibly important on social networks. But like Glinda working in tandem with Elphaba in Wicked (the musical version, of course), the content amplifies the effects of the social cues. With the power of good content behind it, the image becomes even more popular.
Are You Content with Your Content?
Even if you have a lot of followers, though, that doesn’t mean that you should expect a flood of views for any old image you share on Flickr. Content matters, and Khosla uncovered several trends.
“We observe that images with low predicted [popularity] scores… tend to contain clean backgrounds with little to no salient foreground objects, as compared to the high popularity images.”
What makes an image popular?
A. Khosla, A. Das Sarma and R. Hamid
International World Wide Web Conference (WWW), 2014
So, for a popular photo on Flickr, one suggestion is to place an object in the foreground. Makes sense; people rarely enjoy viewing photos when they have to ask, “So what am I looking at here?”
Other observations regarding image content fit just as well with what users have noticed on photo-sharing sites. (Please keep in mind, what follows are largely the musings of a simple blog writer. I highly recommend reading the paper for more specific, scientific insights.)
Color Me Red
Research says this is likeable.
Overall, images that predominantly were composed of cool colors, blues and greens, were less popular than those that contained mostly warmer, reddish colors. As Khosla put it, “This might occur because images containing more striking colors tend to catch the eye of the observer leading to a higher number of views.”
It also might occur because of principles of design and color theory, promoted in materials from such institutions as the University of Texas and Simmons College, that list red as a color of movement and stimulation. Perhaps it stimulates people to click on images as well.
It’s So Fluffy!
Research says this is likeable. And so cute.
Khosla also examined the subjects of the photos and found, to little surprise, that photos of animals generally got a lot of views, as the algorithm predicted they would.
Research has suggested that, in order to trigger the human caregiving instinct, evolution made it so that we’re automatically interested in creatures that look even a little bit like human babies, which means that baby giant pandas and wrinkly little pugs fit the bill. Clicking on photos simply may be one manifestation of the need to nurture. (Research also has demonstrated that looking at cute animals has been shown to improve performance of fine motor tasks and visual search tasks at work, so… please don’t fire me for the pugs.)
Let’s Talk About Objectification
In terms of photo subjects, however, Khosla found that some objects were even more likely to make an image popular than cutesy-wutesy animals, namely, bikinis and miniskirts.
It fits with what Mikolaj Jan Piskorski, a professor at Harvard Business School, observed in 2009. In an article for HBS’s online magazine Working Knowledge, Piskorski noted that, across social media sites, “[t]he biggest usage categories are men looking at women they don’t know, followed by men looking at women they do know.”
It’s just that, on Flickr and other photo-sharing platforms, some of the women the men don’t know happen to be wearing bikinis and miniskirts.
Research doesn’t know what’s good.
Before you get the bright idea to stick pictures of scantily-clad women on your company Flickr account, however, shouting, “Science made me do it!” remember that Khosla’s research takes into account the actions of private users. There’s this whole movement against brands objectifying women that I’d strongly urge you to keep in mind when using Flickr, Instagram, and the like. As author George R.R. Martin reminded an interviewer, women are people, too.
For the sake of comparison, Khosla also mentioned some objects that actually tended to make photos less popular. Among them were spatulas and space heaters. Somewhere out there, though, is a spatula manufacturer with a brilliant photoshoot idea that will buck this trend.
A Totally Unscientific Test
Here are two photos that I uploaded to the QLP Instagram account, which is linked to our Flickr account. Based on what you’ve read so far, which one do you think got more views?
Contender #1: A photo of one of our employee’s two cute kitties that proves that it is possible to get too comfortable with those you share space with.
Contender #2: A photo taken while our Media Team was shooting a product video for the Cloud Stress Reliever.
If you guessed that the cats would come out ahead, you were right. As of May 27, 2014, that photo had 101 views on Flickr (and seven likes on Instagram), while the picture of QLP’s angel got 44 views and five Instagram likes.
However, just for the sake of comparison, here’s a third image we shared on Instagram and Flickr. No color, no kitties, and Bubba’s not wearing a bikini (though he is pantsless):
That image, as of May 27, 2014, had 88 Flickr views, along with 12 Instagram likes (and one Instagram user saying she loved it, d’awww).
So what’s the difference? Well, we could argue that the world loves Bubba (and I’m encouraging use of the hashtag #yesthebubba to prove it). We could also argue that some of the other content factors that Khosla discussed but that I didn’t address, like texture and facial/scene recognition, played a role.
The biggest difference, though, is that Bubba was drawn in pencil in that image. And that’s my point: Khosla’s algorithm might be able to guide users toward popular photos, but there’s also room to do things differently. Want proof? Go search hashtags like #illustration on Flickr or Instagram. Not only will you see how popular art is on picture-sharing platforms; you’ll also get to enjoy some really great work.
That idea of using Flickr in ways that involve doing more than uploading single photos might also help users make the most of Khosla’s research. Remember the second factor, social cues, that also determined how popular an image would be?
In Khosla’s paper, the factor that most strongly influenced whether a photo predicted to be popular actually would be was the average number of views of all of a user’s public photos. In other words, if lots of people liked your previous photos, they’ll be more likely to view your next ones.
You can find all sorts of articles online full of tips and tricks for getting views and followers. They include advice you’ve heard before:
- Tag your photos accurately
- Join groups
- Collaborate and talk with other users
However, to make sure you have a solid foundation on which to place that advice, you need to have interesting content, too.
So you have Khosla’s research. You also have our completely unscientific endorsement of using art on your image-sharing accounts. You have some ideas that you can use to experiment with your own social media pictures.
And if you have pugs, you’ll probably have another follower pretty soon.
Do you use image-sharing services like Flickr? What do you think about what this research suggests? Do you have any photo of adorable dogs on your accounts? Let us know in the comments below!