The success of platforms like Facebook and Twitter comes down to the fact that audiences naturally tend to congregate around social and visual content. Recently, the power of social media as a news driver became apparent when Facebook leapfrogged Google to become the top source for online news traffic, according to analytics firm Parse.ly.
There’s no doubt that social networks, then, are a goldmine for advertisers and it’s no surprise that both Facebook and Twitter took steps recently to boost their appeal with advertisers by introducing autoplay videos, which have become increasingly common and popular for advertisers due to the immediate attention they attract. At the same time, companies can charge higher ad rates on these formats, making it a major moneyspinner.
However, the dangers of this kind of autoplay video, without sufficient monitoring of the content, became shockingly apparent on Wednesday 26th of August when video footage of two WDBJ-TV journalists being shot and killed in Virginia appeared in the timelines of many social media users.
The footage circulated quickly on social media, finding its way on to Facebook and Twitter in autoplay mode. The incident was heightened by the fact that not only was the harrowing and disturbing video repeated over and over again, but also because viewers were not able to stop it. Understandably, this resulted in a social backlash.
With so much video content being uploaded to social channels daily, including 300 hours worth of video on YouTube every minute, this kind of problem is only going to become more prevalent unless firm action is taken now by publishers to monitor video content more effectively. Particularly with mobile video live streaming poised to become the next consumer phenomenon, we can only expect more of this kind of thing.
Here’s the problem: most verification tools rely on the user or metadata to moderate content as opposed to analysing the actual content being viewed, making it difficult to validate user generated content as ad safe with sufficient accuracy. However, the good news is that technologies are now available that can ‘understand’ and classify visual content, both image and video, as well as automate and improve the whole process of ad placement and targeting.
WeSEE technology solves this problem by harnessing state-of-the-art image recognition and machine learning techniques to classify the visual web so that publishers, advertisers, SSP’s, DSP’s and DMP’s have access to vital new visual data streams so to operate much more efficiently.
Leveraging this visual classification technology, WeSEE processes image and video content and splits it into a time-lined story board generating unique layers of visual insights and generates keyword descriptors, as well as turn the content into an advertising taxonomy. The technology is designed to understand every single frame in terms of sentiment and context, making sure that video content meets brand safety thresholds ranging from adult pornographic content through to content containing negative sentiment such as plane crashes, violence – and in the most recent case, shootings or killings.
In the case of the footage from the Virginia killings, WeSEE’s image recognition technology can be deployed within the photo and video platform itself to detect graphic content being uploaded. The approach is to harness modern neural networks that can learn to detect and filter unsuitable videos. A neural net is capable of analysing different information sources such as video stream or static frames from the video and making a judgement based on a combination of all this information. Typically a recurrent neural network (RNN) is used for analysing time sequences.
Beyond neural networks, there are classical approaches that can be used for duplicate multimedia content finding – ie. finger printing visual content can create known bad video black lists. The issue with this, however, is that it requires a human to mark the video as inappropriate at first and only then can an automatic solution be deployed to find duplicates and mark all of them as inappropriate too.
Introducing an automated solution on upload that incorporates both visual recognition technology and brand safety criteria provides a comprehensive understanding of the visual content. This removes the threat of an unsafe image being uploaded before being spotted by a moderator, the web audience or the brand manager. It’s not only a safer solution, but can also be a more efficient one if the site has scale – in this case, particularly with social networks – as there is no need to bring in teams of human moderators.
Of course, this will not help those people who have been left traumatised by last week’s tragic event, but only when this technology is implemented will it help prevent it from happening in the future.
Adrian Moxley, Co-Founder and Chief Visionary Officer, WeSEE
Originally posted What’s New In Publishing 4 September 2015