Machine moderation of user generated content

The advent of Web 2.0 and mobile computing has significantly increased the volumes of User Generated Content (UGC). In order to protect their brand and user experience, websites hosting UGC often prohibit the upload of pornographic images in their terms of use. The task of identifying pornographic material is typically undertaken by a human moderation team that manually reviews each piece of content uploaded or reported by the user community.

Challenges of human moderation

The main issue with human moderation, especially when dealing with image content, is its inability to scale in a cost effective manner. Human moderation teams can easily get swamped by the large volumes of content and backlogs can quickly become unmanageable.  This often leads to a drop in the quality of moderation as the human moderators struggle to keep up. The best solution to this challenge is to support the human moderation team with machine moderation.

Machine moderation

The goal of machine image moderation is not to replace the human moderation team but to make them more efficient. It does this by pre-processing the images with computer vision technology to determine the risk profile of the content. The technology moves the high risk material to the top of the moderation queue to be reviewed by the human moderator for final assessment. This means inappropriate content is quickly identified and removed reducing the risk of users being exposed.

Machine moderation can automatically remove images with a very low risk from the moderation queue reducing the number of images requiring moderation by 50% or more. This translates into direct cost savings by allowing for a smaller more efficient human moderation team.

Image Analyzer

Image Analyzer utilizes advanced computer vision technology with 12 different techniques to reliably identify images that contain human nudity. The technology scans each image based on its visual markers and hence it does not rely on a signature database of pre-categorized images. A high resolution image can be scanned in a matter of milliseconds allowing the technology to quickly process large quantities of images.

First generation image analysis technologies were over-reliant upon inaccurate ‘skin tone’ analysis that was prone to generating high numbers of false positives. Image Analyzer has relegated this element to only a small part of a more sophisticated decision making process that delivers accurate detection with manageable false positives. The technology is even capable of detecting greyscale pornographic images.

Benefits of Image Analyzer

•           Reduces moderation queue by 50% or more while detecting 99% of pornography

•             Reduces costs and increases the scalability of the human moderation team.

•             Allows human moderators to review images based on risk score

•             Reduces the risk of users being exposed to pornographic material

•             Allows the creation of automated workflows based on the risk score

–             Delay image post

–             Challenge user

–             Prioritize moderation