predictionio-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pat Ferrel <...@occamsmachete.com>
Subject Re: UR Query Based on Simulated Events
Date Mon, 17 Jul 2017 17:40:31 GMT
If you are thinking of using some custom method to rank items, look at custom ranking. You
can set the items in the sample query below as the highest ranked and they will be recommended.
You can set a ranking for all items or just the few you think will be helpful and it will
not pollute the event stream. It will only be used in user “cold start” cases where the
user has no events but in other cases will have no effect.


On Jul 17, 2017, at 10:08 AM, Cody Kimball <cody-kimball@pluralsight.com> wrote:

This was to do some predictive analytics, and forecasting to evaluate what would be recommended
for various use cases. This way we can avoid polluting the event store with dummy data and
have to clean it up later as well.

Thanks for the quick response, will look into other means to evaluate potential use cases.

On Mon, Jul 17, 2017 at 11:00 AM Pat Ferrel <pat@occamsmachete.com <mailto:pat@occamsmachete.com>>
wrote:
But it seems you are “simulating” these items, why? how? The fall back mechanisms for
new users are pretty flexible in the UR including custom ranking of items, popular, trending,
hot, random, etc.


On Jul 17, 2017, at 9:54 AM, Pat Ferrel <pat@occamsmachete.com <mailto:pat@occamsmachete.com>>
wrote:

Personalization cannot work for users with no behavioral data. Your app should be sending
events in real time to the EventServer. That way the events below will already be in HBase.
Then you just query with the user-id. Item-sets are meant for shopping carts, not events that
you know about but have not been sent to the EventServer yet. If you know them on your side
of the app—send them immediately to the EventServer, this will work to give results based
on real time events.

If there truly are no events for a user, the below method has nothing to send and so the UR
will fall back to use popular items in recommendations. Item-based recs work fine since there
is only some contextual info in the query like the item that is being looked at on-site.

Getting real time context and events is be the best way to use the system, in which case no
special query is needed.


On Jul 17, 2017, at 9:36 AM, Cody Kimball <cody-kimball@pluralsight.com <mailto:cody-kimball@pluralsight.com>>
wrote:

I am using the Universal Recommender, and I have a use case where I would need to query for
a given user (as if that user had performed certain events, but technically hasn't yet, meaning
the event data hasn't been pushed into the Event Store). 

I originally thought this would be accomplished via itemSets, but the documentation seems
to suggest otherwise. My hope is to be able to pass a list of events with the query, without
those events technically existing yet (nor will be created as a result of the query), and
getting a different result set as a result. An example payload may be:

{
   "user": "userID123",
   "eventList": [{"event":"view", "targetEntityType":"item", "targetEntityId";"targetID15"},
                         {"event":"view", "targetEntityType":"item", "targetEntityId";"targetID23"},
                       {"event":"interaction", "targetEntityType":"item", "targetEntityId";"targetID23"}]
}
-- 
Cody Kimball
Revenue Engineer

 Don't Just Keep Up With Technology. Master It! <https://www.pluralsight.com/>



-- 
Cody Kimball
Revenue Engineer

 Don't Just Keep Up With Technology. Master It! <https://www.pluralsight.com/>



Mime
View raw message