Description - A slider/setting in Recommendations criteria to tune how much the algorithm should explore items in the catalog.
Why is this feature important to you - Our use case is to show some random/less relevant items in a set of User-based recommendations so the they're not overoptimised to what the user has previously browsed e.g. if user only or mostly views pages tagged "Educate me", they can be recommended a few pages tagged "Divert me" or "Inspire me" too (different values of the same attribute/field).
It would also help adoption of Target within the business. Assuring Editorial and Product stakeholders that recommendations are not solely decided by an algorithm so less relevant but still important pages can be recommended would help alleviate concern that recommendations are overoptimised (we are a content/media website).
How would you like the feature to work - In the Criteria screen, a slider similar to Attribute Weighting between 0% and 100% or 0 and 1 like AWS Personalize's:
Current Behaviour - Adobe Target doesn't have this feature but here is briefly how it work on AWS Personalize:
AWS Personalize has a hyperparameter called explorationWeight you can configure to set how much the algorithm should explore the recs catalog. Effectively it adds some randomness into the recommendations so new or less relevant items are recommended, e.g. an explorationWeight of 0.3 would include 3 lesser viewed or less relevant items in a set of 10 recommendations. Adobe Target promotions and Attribute weighting are not really the same.
AWS Personalize doco:
https://docs.aws.amazon.com/personalize/latest/dg/native-recipe-new-item-USER_PERSONALIZATION.html
https://aws.amazon.com/blogs/machine-learning/amazon-personalize-can-now-create-up-to-50-better-reco...