Watson helps AdYouLike screen content for native ads

IBM’s Watson supercomputer is now prescreening content for native ads.

The supercomputer’s newest job is to assist AdYouLike, a native ad supply-side and publisher platform out of the UK. The company, founded in 2011, says it offers the largest in-feed native ad platform in Europe, and it recently set up an office in New York City.

CRO co-founder and US General Manager Francis Turner told me this is the first time Watson has been employed to help target native ads toward compatible surrounding content and determine which content is objectionable to a brand.

Turner said that Watson, accessed via an API, is being utilized “to understand and read content as a human would do.”

For managed service engagements, AdYouLike feeds Watson the brief and KPIs so he knows campaign goals, as well as desirable and undesirable surrounding topics for given campaigns. The advertiser can also use the AdYouLike programmatic platform via a demand side platform, in which case the brand sets up its own parameters.

Then, Watson conducts an analysis in real time of roughly the top 20,000 published web pages per hour in AdYouLike’s network. Turner said the analysis involves sentiment, emotion, semantics and topic categories. There are about 2,000 publishers in the network, each of which has thousands or hundreds of thousands of pages.

Based on Watson’s feedback, the platform can better direct ads to compatible surrounding content, or away from objectionable articles, such as an article about obesity when the ad is for chocolate. (Interestingly, Watson — in his Chef Watson incarnation — knows a whole lot about flavors.)

I asked Turner how Watson’s analysis differs from the kind of sentiment and topic analysis of contextual content that other services provide.

“Watson’s power takes it to another level,” he said, such as increasing the topic granularity. He noted that there are about 30 different content topic categories set up by the Interactive Advertising Bureau (IAB). Watson, Turner said, generates and adds as many as another 40 categories for labeling content.

Previously, he said, AdYouLike’s assessment of content surrounding its native ads was based on searching for selected keywords in that material.