Industry

Online Community & Photo Marketplace

Use case

Automatically flag and moderate unwanted nudity from user-generated content (UGC)

Result

Increased hit ratio for unwanted content by 700x, mitigated Trust & Safety risks, created a better user experience

Photobucket is the one-stop shop for storing, accessing, and sharing photos. Services offered include:

  • Safely upload and store your photos from anywhere 
  • Transform your photos with easy to use editing tools
  • Quickly share your photos with just one click
  • Browse and search for content uploaded by the Photobucket user community
  • Turn your photos into stunning print products

Mike Knowles,  Lead Developer @ Photobucket

Mike is a lead developer at Photobucket where he works on the REST APIs and backend processing tasks that support the front-end web and mobile applications. Mike and Clarifai connected when he was researching ways to apply advancements in machine learning to improve Photobucket’s legacy processes for search and content moderation. Mike is passionate about building great products and applying new technologies in practical ways to improve the user experience.

Challenge

How do you moderate user-generated content (UGC) at scale?

Photobucket is one of the world’s most popular online image and video hosting communities. The platform hosts over 15 billion images, with two million more being uploaded every day. While user-generated content (UGC) is Photobucket’s bread and butter, it also poses a Trust and Safety risk of users who upload illegal or unwanted content. With a firehose of content constantly flowing in, it’s impossible for a team of human moderators to catch every image that goes against Photobucket’s terms of service.

Photobucket needed a solution that would provide a highly scalable system for moderating user-generated content while improving the hit ratio of finding offensive content and the productivity of Photobucket’s human moderation team.

“Before Clarifai, we were only able to monitor 1% of our user-generated content for illegal images. With Clarifai, we’re able to see inside 100% of the two million images per day uploaded to our site, allowing for a better user experience for our online community and our team of human moderators.” – Mike Knowles, Senior Infrastructure Developer

Solution

Photobucket uses Clarifai’s Not Safe for Work (NSFW) nudity recognition model to find and remove illegal and unwanted content from their platform, creating a safer community and better user experience for all 100 million of their members.

Before turning to Clarifai for computer vision-powered moderation, Photobucket used a team of five human moderators to monitor user-generated content for illegal and offensive content. These moderators would manually review a queue of randomly selected images from just 1% of the two million image uploads each day. Not only was Photobucket potentially missing 99% of unwanted content uploaded to their site, but their team of human moderators also suffered from an unrewarding workflow resulting in low productivity.

To catch more unwanted UGC, Photobucket chose Clarifai’s Not Safe for Work (NSFW) nudity recognition model to automatically moderate offensive content as it is being uploaded to their site. Now, 100% of images uploaded every day on Photobucket are passed through Clarifai’s NSFW filter in real-time. The images that are flagged ‘NSFW’ are then routed to the moderation queue where only one human moderator is required to review the content. Where’s rest of the human moderation team? They’re now doing customer support and making the Photobucket user experience even better.

Implementation

Automatically “see” and understand images

Using Clarifai’s NSFW recognition model with a probability threshold of 0.85 (85% likeliness that an image contains NSFW content), Photobucket saw immediate positive results in moderating UGC. When Clarifai recognizes an image uploaded to Photobucket as potentially NSFW, the image is sent to a queue for human review. Upon human review, 70% of those flagged images turned out to be unacceptable content like pornography, while the other flagged images fell in the acceptable range of swimsuit shots and questionable selfies.

Clarifai has been invaluable in improving our hit ratio for finding content with nudity and making our moderation team more productive overall. With Clarifai, we saw a 700x higher hit rate for inappropriate content. Before Clarifai, our hit ratio for NSFW images was something like 0.1% and now it’s 70% .

Protect your community and brand from harmful content

High-quality UGC is important to keep users on your platform and engaged in your community. Bad UGC, particularly in its more harmful forms of exploitative pornography and hate speech, hurts everyone. As part of their Clarifai-powered moderation workflow, Photobucket’s human moderators looked at accounts associated with flagged NSFW images to find even more unwanted content. Within the first month of using Clarifai, Photobucket was able to discover two child pornography accounts that they then passed onto the FBI. Not only was Photobucket able to make their platform safer, but they were also able to make the internet and the world a better place.

“Clarifai has helped mitigate the legal and financial risks associated with running an online community and content hosting business by identifying unwanted images on our platform. But, more importantly, Clarifai has helped us make the internet a safer place.”

Quick and easy implementation

Photobucket developer Mike Knowles was looking for a quick and easy way to implement machine learning-based image recognition technology in his tech stack. After ruling out building machine learning in-house as too costly and inefficient in the long-run, Mike decided using a visual recognition API would be the best way to validate his idea and go to market quickly. He tested half a dozen visual recognition APIs including Google Cloud Vision and Amazon Rekognition before deciding that Clarifai offered the best possible solution for his business.

Mike selected Clarifai based on the superior accuracy and ease of use of the technology, the transparency of the online demo, the completeness of the documentation, and the enthusiasm and professionalism of Clarifai’s team. He was also excited about the wide range of visual recognition models Clarifai has to offer, including the General visual recognition model that recognizes over 11,000 concepts and the Moderation model that currently recognizes different levels of nudity (e.g. explicit and suggestive) along with gore and drugs, with future plans to recognize symbols of hate and violence.

With a product team of four, Mike was able to launch Photobucket’s new content moderation workflow using Clarifai in 12 weeks from concept to internal rollout of the new moderation workflow process. With the new workflow increasing productivity for the human moderation team, 80% of Photobucket’s human moderation team was able to transition to full-time customer support.

“Clarifai has the focus and professionalism of a company that has been in business for ten years. They see beyond the dollar value of their customers and really worked with me to achieve my goals while respecting my budget. Using Clarifai gave me the results I wanted immediately, and their product roadmap gives me confidence that they will provide even more value to Photobucket over time.

DIY with Clarifai

Now that you’ve been inspired by Photobucket’s automated content moderation workflow, it’s time to build your own. Clarifai’s core model includes tags for over 11,000 concepts you can apply to your business. Or, you can use Clarifai’s Custom Training solution to teach our AI new concepts. If you want more personalized insight into how Clarifai’s technology can optimize your unique business, let us know at sales@clarifai.com and work directly with our Data Strategy Team and machine learning experts!