predictionio-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pat Ferrel <...@occamsmachete.com>
Subject Re: how to embed user-user relations into UR engine
Date Tue, 22 Nov 2016 15:38:21 GMT
Anything tied to a user that you think may indicate their taste for a conversion item can be
encoded in secondary events and are tested for correlation to the conversion on every conversion
item. These “secondary events” do not have to have ids that can be converted on, so (user,
“member-od”, group-id) is a perfectly fine secondary indicator event.

I use the term conversion or primary events to mean the thing you want to recommend on the
item type you will recommend. So the conversion may be a purchase on a product-id but it may
very well be useful to train with a user’s membership in some group.

BTW this is the greatest strength and the most innovative part of the UR. It can recommend
something a user may want to “purchase” even though the user has not seen any products,
just by looking at other patterns of behaviors like group membership. These secondary events
can be very strong (share is a strong indicator of taste) or weak (gender is hard to make
very effective) but there are ways to measure this.

On Nov 22, 2016, at 4:05 AM, Guruju, Lakshmi Sravya <lakshmi.sravya.guruju@sap.com>
wrote:

Hello everyone,
 
I am building a recommender using UR engine template. As far as I know, this usually takes
events like view, buy etc. These events mostly tell the relation between user and the item
as in user viewed an item or user bought an item.
 
But in my use-case I have some user-user relations like set of users belong to a specific
group which I would like to somehow import as an event to event server so that they can be
used to give better recommendations by the engine.
 
Could someone help me in achieving this or suggest me if there is a better way to use that
relations in getting recommendations.
 
Regards,
Sravya


Mime
View raw message