predictionio-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pat Ferrel <>
Subject Re: Use user property for recommendation
Date Wed, 03 May 2017 15:57:02 GMT
This is done automatically since we know the user’s history of purchases. In fact they may
have many tendencies, not just one and the UR also compares these to other users to find what
similar users bought. This is the essence of collaborative filtering, finding from the data
what tendencies the user has.

The UR has flexible business rules that work like you mention below but for tendencies, leave
it to collaborative filtering—that is what it was invented for.

For instance, you may have a “buy” event as the conversion event, and another “category-preference”
event that is fired when a “buy" happens. This event might be encoded (user-id, “category-preference”,
“toy”). The UR will use all the category-preference events, determine which led to “buy”s,
compare them to other similar users, and conclude which items have an affinity for people
with a category-preference of “toy” then at query time if the user has a category-preference
of “toy” in their history the best items will be ranked in recommendations accordingly.
It may be likely that the user has more than one category preference so we choose the best
items using all indicators, other buys, other category-preferences, etc.

On May 2, 2017, at 10:53 PM, Lin Amy <> wrote:

Hello everyone,

I would like to use user property for recommendation and need some help, since it doesn't
seem available with current version of "universal recommender".

For example, say user A has property "tendency": ["toy"], and the user A is specified in the
query, so I expect the engine to recommend items with "tendency": "toy" by given bias.

Is there any way to do this?
Thank you so much for any advices.

Best regards,

View raw message