What methods are used in Neuwo’s text analysis and what makes it so accurate in the content classification? We uncover Neuwo AI and how to successfully implement content-driven solutions.
There are many different methods that can be used to analyze text. We are often asked whether Neuwo technology relies on the existing solutions offered by Google, Microsoft or Amazon. We don’t – there were no ready-made solution that can offer what we wanted to accomplish. Base models for simple classification and clustering exist but they are based on common ontologies. Our customers’ needs are often specific, relating to particular themes or regions, so Neuwo’s ontologies are built according to these needs.
Even one unmapped category can make a big difference
The requirements for quality on content classification are very high, because even one undetected ad category means no ads are displayed. Accurate taxonomy mapping, i.e. assigning the taxonomy categories to content required new types of models, so we built our models from scrap. The IAB-Taxonomy mapping we created at the Tier1 and Tier2 levels reaches an F1 score higher than 90%; this determines the accuracy of the classification, reflecting the amount of time and data spent on buiding our technology.
BrandSafety needs to react to changes in real time
It’s the media’s job to pick up new trending topics as soon as they happen. Russia offensive strike in Ukraine set an example of how quickly machine learning models should be able to adapt to an evolving news cycle. Brands have a valid concern to be careful around articles that are particularly negative, such as those covering war, famine, or death, and they need to able to target their advertising to suitable topics.
However, advertisers play an essential role in supporting quality news content. Even in times of crisis, advertisements can still appear in, for example, financial news, such as articles on economic sanctions, articles on general security policy and aid measures in various countries. Neuwo’s trend tracking based on phase differences and multi-layer modeling enable near real-time response to such changes.
Successful similarity multiplies the browsing time
The distance evaluation of Neuwo’s articles – meaning how we find similarities – is the result of our collaboration with the customer on modeling. When algorithms focus on finding genuine similarities in content, not just detecting the same keywords, as the result we can offer the reader content that makes them stay on the site longer.
Contact us to schedule a demo to learn more!