Navigating the wild world of contextual – and what to look out for
As part of The Drum’s week-long Data and Privacy Deep Dive, GumGum’s Ken Weiner spells out how advertisers can tap into the growing promise of contextual advertising in light of signal deprecation across the web.
/ Adobe Stock
Contextual targeting is as old as modern advertising itself. Long before the now-endangered third-party cookie emerged, ads were often targeted based on the content they appeared beside – think of an airline ad appearing in the travel section of a newspaper.
Now, as the industry is shifting away from audience targeting due to growing data privacy regulations, advertisers are turning back to contextual targeting to help find the most optimal environments to place ads without tracking people or using their personal data.
Contextual targeting is now a staple in most adtech providers’ offerings but, like many technologies, not all are created equal. The majority of contextual solutions are still just basing their contextual understanding off of keywords alone, which we’ve learned gives a very limited understanding of content and whether it’s safe and relevant to advertise on.
Enter the brave new world of contextual intelligence, an era in which all of these creases will be ironed out. AI advances now mean automated contextual analysis has attained a human-like understanding of digital environments across all platforms or channels. But not all contextual solutions have reached this level of sophistication – here’s how advertisers can tell the difference.
Has it mastered natural language processing?
The limitations of most keyword targeting solutions are borderline funny – if they weren’t so serious. We’ve seen instances where these tools have identified an article about ‘classic hi-fi systems’ as pornographic simply because of the letters ‘ass’ in the title. For similar reasons, an online recipe for ‘killer key lime pie’ would likely be blocked as potentially violent content.
Natural language processing (NLP) – a technology at the vanguard of contextual intelligence – is changing all this. NLP allows contextual tools to analyze not just individual keywords but all of the words and letters on the page. With this information, systems get a better understanding of the nuanced meaning of language. This goes right down to comprehending the cultural nuances of individual words – for example, ‘killing it,’ when used colloquially, generally describes success or brilliant execution. NLP is now broadly applied to audio content too, allowing advertisers to analyze the voiceover within the digital environment, including podcasts.
Does it have computer vision?
NLP is bringing a level of nuanced understanding to language analysis that should now be considered standard in the contextual space. But even NLP isn’t enough on its own. After all, the online world we know so well is just as, if not more, dominated by images and video content.
Luckily, contextual intelligence is mastering this type of content as well, with AI-based computer vision allowing systems to comprehend images at a micro level, scanning every color, pattern or object to understand its meaning and threat level. Going even further, CV is able to analyze videos on a frame-by-frame basis and know exactly what a video is about and which ones offer the best opportunity for ad placements.
Does it reach CTV environments?
The ability to scan video with such detail and clarity has obvious applications for the burgeoning connected television (CTV) and over-the-top (OTT) media space.
It’s an area that has grown so quickly that brand safety technologies are still catching up. Most advertisers, when targeting their ads for CTV, are still relying on basic video metadata. But, like keywords with text-based content, metadata alone tells you very little about the nature of a video and the dangers it contains.
With computer vision, contextual targeting is able to go much deeper and fully understand CTV content second by second, avoiding dangerous placements and freeing up safe inventory. If a video contains violence, metadata will often designate the entire video as unsafe, but deeper analysis could find that the violence in question is limited and only appears at a certain point, allowing intra-video metadata to be applied and indicate exactly where ads should be avoided.
Is it accredited for content-level analysis?
Viewed as a whole, these emerging contextual capabilities now allow us to analyze all of the key digital content signals, whether text, image, video or audio based.
Keyword analysis sits in the more basic property-level analysis because it only really deals with language and not even in the way it should. That’s why advertisers must be seeking out contextual solutions and partners that are properly accredited for content-level contextual analysis, as it becomes the standard for the technology.
With these advances, contextual intelligence is now entering its most exciting phase yet, becoming one of the pillar technologies of the digital world. We like to see it in these terms: if property-level contextual was black and white, content-level analysis is the moment the space moves into the technicolor age.
Ken Weiner is the chief technology officer at GumGum. For more on how the world of data-driven advertising and marketing is evolving, check out our latest Deep Dive.