✨ New ✨ The Digital Services Act: A fireside chat covering all angles Watch it here → ×

How Computer Vision Can Improve the Quality of Your Content


    When we think content we think words first of all, right? But, as the adage goes, a picture speaks a thousand words — which is particularly relevant where content moderation’s concerned.

    Online marketplaces (like classifieds sites and dating sites) pretty much rely exclusively on user-generated content. Although moderating text using machine learning algorithms is pretty common, images — which present a composite mix of criteria — have been harder to vet using AI. Until now.

    Computer vision – amongst other things used for image moderation — is becoming powerful enough to change that. The ways it’s evolving across tech are numerous, but in terms of content moderation, it serves a very specific purpose: it can automatically review, approve, or reject digital images and video.

    Algorithm-powered image moderation can distinguish between any number of image features — from obvious no-nos like pornography to more under-the-radar naughtiness like contact details embedded into a photo. And of course, it can weed out accidental flaws and faults like low-quality images and duplicates.

    Let’s look at what you can moderate with computer vision…

    Recognizing human features

    Not all online marketplaces want pictures of people on product images. They take away focus from the product and can lower the user experience. Computer vision can spot and, if necessary, remove pictures featuring people rather than goods.

    However, on other online marketplaces, like dating portals, faces are pretty essential. It has to be the right faces, though! For profile pictures, you only want the face of the account owner. Group photos will only confuse, and that’s where AI image moderation comes in. Computer vision can reject pictures with multiple people in them, selecting only pictures that show the person looking for love. It can also screen for unsavory images or pictures where faces are obscured by objects and determine things like gender, age, and skin/hair color, which it can cross-reference with accompanying text. Bad news for those still posting their prom photo 20 years later. Time to update your photo, so it matches your details…

    Offensive images

    This kind of goes without saying, but it’s good to know that the really bad stuff gets nixed straight off the bat. And we’re not just talking about sex or nudity: with the right data, computer vision can be trained to screen for violent and shocking images, as well as things like religious or political symbols, which can be used in negative contexts.

    It can also tell the difference between something mildly suggestive in an image — like cleavage — from full frontal nudity. Ultimately,  users will lose trust in a site if they encounter really nasty stuff, which is why image recognition is so important. It also helps curb user churn by vetting contributed content and ensuring it lives up to the quality and standards expected by the users. 

    Image text and watermarks

    Some online marketplace sellers will try to circumvent platform’s sales fees by posting their contact details in an image to encourage off-site transactions. Computer vision stops this practice before it starts by locating text and watermarks in images, preventing sellers from linking to their own (or competitor) sites and driving traffic away from your site.

    On dating sites this technology can be applied to ensure that contact details are not shared outside the channels endorsed by the site owner to secure revenue and to keep users safe.

    All aboard the computer vision train

    Creating and running a successful online marketplace comes with a lot of challenges. Your main focus needs to be on continuously expanding and improving your core product to stay competitive, meaning most companies don’t have the time or resources to moderate all content that’s submitted manually. Even if they did, with a 100% manual setup, they’d struggle to meet user expectations, as content wouldn’t instantly go live. This is where automated moderation can make a big difference. It’s cost-effective and allows for quick content upload. Another benefit is that it enables data privacy in a way human moderation can’t, which becomes more relevant in the digital world.

    As you can see, there are plenty of reasons to jump aboard the computer vision train, but it takes dedication to get AI-powered image moderation right: unless you have the right help.

    Computer vision is a core component of Besedo’s service. If your site handles user-generated content and accompanying images, then take a closer look at how we can help.

    This is Besedo

    Global, full-service leader in content moderation

    We provide automated and manual moderation for online marketplaces, online dating, sharing economy, gaming, communities and social media.

    Form background