‘CBC generates almost 500 new pieces of content each day so we need a system that can make real-time predictions and be updated with new pieces of content as they emerge.’
Canada’s public broadcaster is refining a new form of recommendation engine based on session preferences. Many recommender algorithms predict interest by either by
- recalling a user’s prior searches (‘you liked that song so you my like this one’), or
- inferring interest by similarities with searches by others (‘other people that liked this book also liked these’).
The per session approach predicts a trajectory of interest based on the items selected earlier in the visit.
The Public Service Algorithm: Personalization at the CBC
MEDIUM | October 1, 2018 | by Jason Cornell