Illuma, the contextual-AI specialist, has developed a new content classification system which may soon improve accuracy to as much as 100%*.
The Illuma data and insights team has spent four years turning its similarity engine, which was developed in 2014 to expand campaigns and boost performance, to the task of improving the accuracy of contextual targeting.
This new approach to content classification combines the scale and speed of Machine Learning with the intelligence and accuracy of human correction, using a branch of AI called Human in the Loop (HITL).
Illuma CEO and founder, Peter Mason, said: “As the industry has evolved from basic keyword classification towards Machine-Learning models such as NLP, the accuracy of contextual ad targeting has improved significantly.
“However, machines will never be able to compete with a person when it comes to reading a piece of content and explaining what it means. While using humans in web classification is not new, using them in this way, for this purpose, is breaking new ground.
“In 2020, Illuma embarked on an ambitious project to humanly classify CTV and web content in order to fine-tune our Machine-Learning models until they delivered almost 100% accuracy in tests. By deploying HITL, the Illuma data and insights team has set new standards for accuracy and addressed one of the key pain-points of contextual targeting.”
Instead of using its contextual recommendation engine to expand reach and improve performance – the original purpose of Illuma when it was developed in 2014 – the technology is now being used to find near-identical matches to a humanly classified ‘truth source’ across millions of pages or videos.
As part of the development, Illuma’s HITL contextual categories across web and CTV have been updated to cover 400 IAB v3 categories, greatly improving the granularity of targeting choices available.
Peter Mason said: “The online advertising industry has always had to accept a degree of inaccuracy and therefore wastage with contextual targeting. We wanted to leverage our AI to try and improve this situation.
“Illuma is now using Machine Learning to do something which keywords and NLP have been doing for some time with mixed success, and allowing for the scaling of highly accurate, human-categorised pages across the web and CTV.”
As well as solving real-world problems for marketers, Illuma’s new categoriser represents a step forward in the way AI is currently used for classification. The findings are also of academic significance and interest which marks a real achievement for everyone involved.
Illuma head of data and technology, Dr Yu Liu, said: “The Illuma categorisation engine is built on a combination of human classification, intelligence and correction – the highest level of accuracy – along with machine interpretation and classification, which brings almost-unlimited scale. This is an interesting development in the field of AI.
“It also allows for improved accuracy at very granular levels regardless of the quantity of input data, and can output into any chosen classification; this could be the latest IAB v3 or a proprietary set of classifications.
“This is exciting news for DSP and SSP platforms, brands and agencies looking to target with precision across a range of different environments with varying contextual data available as inputs.”
* Allowing for subjectivity
