Slide img img img images

Our patent-pending


Method

We’re the most precise and transparent
contextual advertising solution for programmatic display.

Why better contextual advertising?

Sentiment analysis and entity detection don’t work well for the precise identification of news content. For instance, take this article about George Floyd and the influence his death is having on affordable housing in D.C. Current brand safety technology would label content as having the following sensitive topics:

They're not wrong, but they're missing the social context.

Hands-off approaches fail to capture the big picture (what we call the social context). While this article may not be for all brands, those that celebrate racial equality would be remiss not to consider it. Yes, leading brand safety solutions almost certainly block this content. Why? Because they use unsupervised methods — that is, methods where experts aren’t at the controls.

Unsupervised ML

Existing tech is good at “binning” content based on keywords. An unsupervised model is good at learning that the words “shooting,” “gun” and “violence” are related and tend to appear together in news. This is better than manually selecting keywords to block, but it’s hard to understand how exhaustive or precise it is at catching content.

Our Supervised Deep Learning Process

The socialcontext solution

iconfinder_bias-unethical-prejudice-cheat-unfairness_5811902
Define the filter or segment

We start with a rigorous definition for each news filter and segment we build, based on sound academic research. We’re experts in the precise analysis of news content.

iconfinder_detecting-research-news-check-verify_5811874
Capture examples

We capture and identify thousands of examples. We debate over questionable articles and edge cases to ensure each article is a good match.

iconfinder_consider-source-detecting-research-check_5811897
Generative AI

We custom design large language models to precisely label content. We regularly audit and fine tune models to ensure they understand the problem at hand.

iconfinder_fact-truth-news-right-clear_5811875
External validation

Each algorithm is validated. Before it’s ready, it must perform well on news it has never seen. Model updates ensure they stay accurate.

Unparalleled specificity

We design our filters to block specific types of content advertisers find sensitive. For instance, our anti-vaccine news detection filter specifically catches bad content and leaves the good alone. Unsupervised methods don't offer this precision. We go beyond topic and sentiment detection to detect the specific problematic linguistic fingerprints of bad news content.

Socially equitable

Brand safety floors only take inventory away from advertisers and publishers alike. Too often, this process eliminates good news content. Socialcontext knows what’s bad, but we also identify the good. That is, our AI detects socially equitable content. Our pro racial-diversity filter unblocks five percent of sensitive news content alone. This means more revenue for news publishers and more inventory for advertisers.

Accurate and transparent

We validate each news filter to ensure it’s good enough before you use it in your ad tech stack. That means that each news filter we build performs well on news it has never seen before we ship it. In an effort to be the most transparent BSF out there, we share these performance evaluations with our clients, so they can reasonably understand how well their brand safety floor is performing. We’re never satisfied with our models. All filters get periodically updated with new data to make sure they stay accurate across time.

MORE INFO

Learn more about how we are improving contextual advertising

CONTACT US

Interested in reaching out?