‘Filthy, Sexy Pasta’ Isn’t What You or Keyword Analysis Thinks It Is

Here’s how to avoid contextual pitfalls

In a post-cookie world, the stars are aligning for contextual advertising and a host of technology providers are emerging to take advantage. But not all of these contextual challengers come with the same expertise.

The tech that allows computers to read the content within a web page before placing a relevant ad in real time is complex. While some providers will have mastered the advanced capabilities required, others will be relying on more rudimentary technologies.

Take keyword targeting for example. Deployed on its own, keyword analysis is actually a very blunt tool. Just look at the millions of perfectly safe pages on Covid-19 and Black Lives Matter that were needlessly blocked last year because keyword analysis wasn’t able to identify genuine threats.

Thankfully, advances in machine learning are opening up a more nuanced and sophisticated approach to contextual targeting. But it’s up to advertisers to ask the right questions so they get the very best this tech can offer. Before selecting a contextual partner, here are a few key questions to ask:

Does the contextual engine deploy natural language processing?

Some of the errors that occur with keyword blocking would be almost comical if they weren’t so serious. Killer Key Lime Pie and “Filthy, Sexy” Pasta Sauce recipes are just two examples of innocuous pages that would be miscategorized as violent or pornographic by keyword blockers, simply because of a single word in their titles.

To avoid mistakes like this, ask your contextual provider if they deploy natural language processing (NLP). NLP is a branch of machine learning which allows computers to understand the text within a page as the human brain can, picking up on the nuance and broader context of language. Armed with NLP, a contextual solution would understand that a killer key lime pie or filthy pasta page is actually food content and could be a valuable environment for a retail or restaurant ad.

NLP should also be used by providers to analyze the spoken words within voice overs or audio on a page, in order to gauge brand suitability. Audio is a blind spot for many contextual solutions.

Can the technology understand the full context of a page?

Beyond text and audio, contextual providers also need to incorporate image- and video-based data signals if the full context of a page is to be understood. Computer vision (CV) is the visual equivalent of NLP, giving computers human-level understanding of images and videos, but on a larger scale. Most contextual solutions still rely only on analyzing the metadata of video alone, which often tells you very little about the content.

For example, metadata analysis would struggle to identify hate iconography within video content because that red flag may not appear in a video’s metadata description. On the flip side, such basic tech could also unnecessarily block a video about a killer key lime pie recipe, because it reads the word “killer” without further understanding the video’s innocuous context. CV is able to look at the objects and shapes within visual content itself on a frame-by-frame basis, giving a more precise reading of brand suitability and safety on any given page.

Is the contextual provider accredited by an industry body?

With so many new contextual solutions and challengers entering the market, advertisers need robust industry standards with which to validate potential partners. Thankfully, the Media Rating Council is taking a lead in this regard.

Currently, the status quo for contextual accreditation is property-level: This means a provider has mastered the analysis of text-based data. But to ensure contextual awareness, advertisers should be looking for providers like GumGum that have earned content-level accreditation, meaning their solutions incorporate all available signals (text, image, audio and video) within a digital environment, rather than just text alone.

Put the provider’s skillset to the test

Your provider should be able to share case studies and results from other clients to demonstrate its contextual capabilities. But to take this a step further, advertisers shouldn’t be shy to put the vendor’s skillset to the test themselves. One relatively simple way to do this is to create a spreadsheet of 1,000 web pages and include a dozen or so pages that you know would be relevant for your brand but would most likely be blocked by keyword analysis.

For example, a fashion retail brand could perhaps add a page on sexy summer dresses. If the provider flags a page like this as pornographic, you will know that the system it is using is rudimentary and unlikely to understand the full content of the page.

In a post-cookie world where advertisers must walk the line between consumer privacy and brand reach, contextual targeting is emerging as something of a super vessel by which to navigate new waters.

But players across the digital ad spheres must be proactive in holding contextual suppliers accountable. Are they really the trailblazers they appear to be or does their tech lack the residual substance and nuance required for a complex, noisy digital world? It’s only by answering these questions, and really scrutinizing the machinery on offer, that brands can separate the pretenders from the true captains of the industry.

William Merchan is a data science, marketing analytics, advertising technology and startup veteran. He currently serves as head of Verity for GumGum, where he leads the artificial intelligence platform.