Using properties of items there are new exclusion rules is in the next release 
of v0.6.0, which will be released shortly after PIO v0.11.0 and Mahout v0.13.0, 
both are going release now.

“Bias" in The Universal Recommender v0.6.0:

0: exclude items with properties specified (must_not in ES parlance)
<0: include only items with properties specified (must in ES parlance)
>0: use the bias as a multiplier on all scores, >1 should be considered a 
>“boost"
However item based whitelists will completely ruin the performance of a 
recommender. They are implemented now when you set in inclusion field in the 
query of “id”: [“id1”, “id2”< …] This is undocumented because it is a bad 
idea—read on.

I warn you that this will restrict recommendations so much that it will not be 
worth making recommendations. You are telling the recommender to go through a 
bunch of work to create the best recommendations then throw them all away and 
rank the few items you have whitelisted. It may be that none of the whitelisted 
items are able to be recommended to begin with so you may get no recs.

If you simply want to rank things on your whitelist, use popularity or 
something that will be more likely to give you recommendations and even then, 
this will often lead to conversion lift of 0 so be careful.


On Apr 15, 2017, at 6:25 PM, Lin Amy <linam...@gmail.com> wrote:

Hello everyone,

I am using the universal recommender. And I wonder if the following situation 
can be realized by the engine.

First, an item can be set with either whitelist or blacklist property (but not 
both), and the elements in the lists is something like category.

The recommendation will only contain those items with  from whitelist if 
whitelist property exists; on the other hand, if blacklist property presents, 
then those with property from blacklist will never be in the query result.

Is this possible? 
Thank you all!

Best regards,
Amy

Reply via email to