2 min readMost of today’s recommendation engines rely on curation and supervised machine learning. More advanced engines rely solely on reinforced machine learning. We b elieve a hybrid approach integrating all of the above (and more) should be used provide an optimal audience experience and increase engagement and time spent with content.
The majority of practical machine learning uses Supervised Machine Learning. It maps input (X) and output (Y) variables and algorithmically learns mapping from X to Y. Like having an answer sheet to a test, supervised ML provides answers that are accurate for some, or even most—but not all. In the context of a media recommendation engine, this results in the semi-accurate, but often inaccurate—and therefore, impersonal—suggestions we see from most engines today.
On the other hand, reinforced machine learning uses a process of learning through trial and error. Monitoring, prompting, and then rewarding the viewer for exploring and discovering new content.
In a hybrid engine approach, both supervised and reinforced ML are integrated with O.C.E.A.N. (Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism) personality trait data, and cable/satellite first party data, as well as other data sources. Doing this ensures that a media company truly understands the unique profile of each viewer. This promotes recommendations that go beyond curation and viewing history to empower recommendations the individual may have never considered, but in all likelihood will love.
A recommendation engine has truly arrived when a viewer no longer feels there’s nothing good to watch, but rather, struggles with choosing what to watch first because every suggestion is a favorite.
To learn how SoftServe is developing a hybrid media recommendation engine that can be deployed anywhere and integrated with both existing engi nes and new builds, read “THE ROAD NOT TAKEN IN MEDIA”.