In another post we took you through what our particular definition of curation is, why we curate as we do and what we think curation is and what it isn’t.
Our focus in this post is in explaining exactly how we determine the relevance of individual content pages when we curate for brands and how it is different from other offerings. Providers of contextual curation focus on identifying context using keywords at the domain level. Some of the other providers will work to separate news pages from other content. Others will provide a directory of domains which should deliver a similar audience. There is some page level targeting. Our unique innovation was to build the ability, in real time, to ascertain both the topic of the page and the keywords that appear on the pages.
As any search engine marketing professional will tell you, Keyword targeting is a tricky proposition. Just look at the seemingly endless lists of keywords, positive and negative, that lie in any Google or Bing search account..
We don’t think that sort of hand to hand keyword management is practical in a fully programmatic environment. In programmatic you’ve got scale volume to deal with and very little time to deal with it. With too many keywords you just can’t define and deliver relevance in milliseconds.
Also, relying solely on keywords can lead to unintended results. As an example, how do you treat content that contains perfectly good travel keywords but that are on a page about people trying to escape a war. The words themselves are right. But since the topic is people escaping a war, they have nothing to do with leisure or business travel.
Our invention was to run two processes on any page of content in the available milliseconds; one to understand the topic of the page and then a second process to understand the keywords in context. Combining these two concepts into one listening allows our travel brands to avoid keywords that look right but aren’t, given the topic of the page.
We check page topic first and keywords second,at the page level URLs. The URLs represent the content pages that sets the context for the reader; it’s where all the relevant signal is. We don’t rank and score home pages or channel fronts since they’re always changing. A health channel front might have hospital links in the morning and how to get a better night’s sleep in the afternoon. Article pages have stable content where a robust topic model can identify the topic.
With that done then we look at the keywords and we’re able to understand that a page about vacations should be leveraged for travel brands. The travel keywords in news content about war should be avoided. And we do this 1 million times a second across 3.5 billion URLs that sit in our index.
From there we can then assign a page to a segment. When we do that we put the page in the segment and we also note if we’ve seen an ID visit that page in the last 30 days. Now the brand has a choice they can make; target the inventory then or the ID later. We make both the list of pages and IDs available in any digital DSP. We make the IDs available in social and ship to the brand’s ad account for them to target.
Comments