The next big frontier for anyone involved with photography online is content curation. The value of a company dealing with photography is not in its ability to attract a large amount of content – that is the easy part – but rather in its ability to create value to the content by carefully and appropriately curating it.

The first phase of photography content valuation was to amass as many assets as possible. Companies success rate was measured in how quickly and reliably they could accumulate  a vast amount of photos from their users, regardless of their quality. As Flickr quickly realized, that strategy was doomed to fail as not all content is equal.For visitors to keep on logging in, they had to add efficient filters that would hide the less desirable items and make the most popular one bubble up. Their algorithm, called interestingness, relied on user comments, likes, clicks to create an automated ranking of the best images. And it worked, for a while at least.

The issue, after a while, is the necessity to go beyond popularity as the primary trigger for leadership. Popularity triggers more popularity which in turns shuts down discovery. In other words, the more an image is popular, the more it becomes more popular. And as we have seen many times with viral photos, meme or videos, the reason for popularity might  be far from any esthetics reasons. In turn, curation had to evolve.

While it is not that hard for a human to quickly select good images from an incoming feed, it is just not scalable. It works for stock photo agencies like Shutterstock or Getty Images but when your feed gets  million of images a day, like most social media site, it is just not a desirable option.

Automated curation based on aesthetics is still far from being helpful. Back in 2005, researchers at the University of Pennsylvania  created Acquine, a project that learned from people’s votes to replicate human curation. The result was somewhat promising but tended heavily towards the beach at sunset with a boat at the forefront photos.

One of the biggest issue to solve in automated curation is in anticipating the viewing audience. A 20-year-old American male will not have the same interest as a 60 year Indonesian woman. They will obviously not like the same photos. In short, photo curation needs to deliver on expectations by predicting the audience. If you know your audience, it becomes much easier.

Take G+ for example. They know that the majority of the images they receive are family/friend photos to be shared with family/friends so here is their algorithm :

• Removing the blurred images or otherwise poor technical quality.
• Selecting images with faces/people. Recognises the people in your circles as people who’re important to you.
• Picking images that have smiles.
• Removing duplicates (or near duplicates), & getting the best one based on the above.

Simple enough? The same goes for specialized companies like Chute or Curalate that search for specific content for specific brands. They can quickly skim through millions of Instagram uploads and retrieve those that are particular to their needs based on a logo and hashtag for example.  Manual editing can then take over to filter out the false positive or useless ones.

Pinterest, Instagram, and down the line Facebook, Twitter, Google and Yahoo understand that it is no longer sufficient to show images. The images seen have to be relevant in some manner to the viewers. Facebook and Twitter have their work cut out for them as the relevancy of the images depends on who is sharing it, which is defined by the users themselves. Pinterest, however, needs to constantly tweak its discovery algorithm since everything is public and relies entirely on relevancy of content. Imagine what would happen to Pinterest if we started to see mostly blurry images or porn. Same for Instagram.

While big improvements have been made a lot still has to be done. The value of any social media site is only as good as its curation algorithm. As the accumulation of data around image usage, and more importantly, image conversion rate – increase, we can expect to see more potent filtering to a point where every site we visit will only show images we love or want to click on. Which, after all, might dramatically alter what we love.