With user-generated content, lines will always be crossed. We all wish content moderation didn’t have to be a thing, but it is. So our Data Science team created a scalable solution for it, one that we’re able to build and integrate for any client. Once implemented, it’s entirely free, making it 95% cheaper than other options.
Our model is rooted in state-of-the-art AI and the newest ML research that allows it to run on a device and reach the same accuracy as industry leaders doing the moderation in a cloud — something that was impossible just months ago. And by being on-device, it’s the only feasible option for live streaming.
In short, we’re making a generally inaccessible technology available to everyone.
While our ultimate goal is to offer this on-device solution as a standalone service, right now it’s available exclusively for our partners. We’re already seeing great results — more on that below.
An Overview of Available Content Moderation
There are three types of content moderation: manual, backend and on-device.
- Manual content moderation is done by a team of people who label content, which is slow and inefficient. With other options, it’s still necessary to have some manual moderation (for example, with content reported by users and some edge cases), but backend and on-device moderation reduce this need to an absolute minimum.
- An ML model that runs on the backend as a cloud-based service — like AWS, Hive and Clarifai — is faster and, due to limited options, remains the go-to choice for most businesses. Disadvantages include high price, poor scalability, immense limitations with video and security threats.
- On-device moderation is the most reliable option and the only viable way to moderate video chat and live streaming (when done right). However, it’s out of almost every company’s price range — which is why only companies like Apple, Facebook and Samsung have been able to do it — and attaining a backend level of performance was impossible just a few months ago.
The Growing Benefits of Content Moderation
User-generated content (text, images, video) is at the nub of countless platforms, namely social media, dating apps and marketplaces. Keeping it in check based on pre-set guidelines is key.
The App Store and Google Play require content moderation in all apps with user-generated content. Any platform that is accessible to children must have it as well. On top of that, one in three consumers will abandon a brand they love after one bad experience.
Then there are lesser-known benefits, like improving online visibility. Statistics show that 25% of search results from some of the largest brands come from links to user-generated content. This is great news — if you’re sure that the content won’t harm your brand.
All About the STRV On-Device Model
Just like Facebook and Apple have their own content moderation, so can our clients. As a part of our services, we’re able to develop, train and integrate a custom content moderation model smaller than 3MB which can be embedded to any mobile or web app.
Price
As soon as we integrate the model into your app, it doesn’t cost you anything. Content moderation becomes free. In effect, this means that the more content, the larger the savings. For a Twitch-sized company, savings can reach 99.99%.
Scaling
Scalability is infinite because there is no need for a cloud or backend ML model. Our solution scales naturally with each device because the content moderation occurs on the user’s device. As for video content, our model can be scaled based on video length, quality and quantity.
Security
Sensitive content isn’t sent to a third-party service, which negates security threats. The user’s data is processed on his/her own personal device; all content stays with the user.
Speed (crucial for live streaming)
Because there’s no need to upload large files to third-party servers, our model comes with instant processing speed. Results are handled immediately on the device.
With a backend solution, there’s a wait time as the service goes through processing, handling results and reporting back to the device — making it unusable for video chat and live streaming.
How, Exactly, Does It Compare With Other Options?
Let’s say your platform has 10,000 minute-long videos that require moderation. How long does it take each of the content moderation solutions to label them, and how much does it cost?
With a manual approach, one minute-long video takes roughly one minute to moderate. 10,000 videos come to 166 hours, which is one month of work for one person. The average monthly salary of a content moderator in the US is $3,000.
Using backend moderation like AWS or Clarifai, moderating 10,000 minutes of video would take several seconds per video and the overall cost comes to $1,000. This option also comes with an additional risk: unknowingly tying your business to illegal content. If a user shares such content (like child pornography), that content is being sent to a third-party solution over the internet. With an on-device solution, the content doesn’t leave the user’s device.
And finally, having an on-device solution already integrated, the results are immediate and cost you nothing.
Filling a Huge Gap in the Market
Clients need content moderation, users want it, app stores require it. The content moderation space is predicted to grow by 61% in the next five years and hit $32 billion by 2031. Last year, Hive raised $85 million on a $2 billion valuation, and Clarifai raised $60 million.
It was an easy decision to utilize our expertise to create a solution that’s attainable for everyone. And because our solution provides the only viable way to moderate video chat and live streaming, it’s an incredible competitive advantage for our clients.
Right now, we’re implementing it for numerous clients, such as Cosmic Latte — a group of leading dating apps (including Zoe and Surge) that has more than 10 million users and facilitates about one million matches every month. We can’t share the details of other projects as they’re in the development phase, but client feedback has been positive and we’re thrilled to see the results so far!
