predictionio-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pat Ferrel <...@occamsmachete.com>
Subject Re: E-commerce recommender, recommendations for users who are not logged in?
Date Mon, 20 Mar 2017 16:32:56 GMT
Yes, you can filter by category if they are attached to items as properties. In 0.6.0, soon
to be released, there are inclusion, exclusion, and boosts by any property that can be attached
to items. Blacklists are for item-ids only, either items that have gotten certain events or
just ids. The business rules are pretty flexible.

BTW if you have small enough data you can run with Postgres + Elasticsearch. Be warned that
though I maintain the template I don’t have a lot of experience with Postgres.

Something to consider is that moving from most web use cases to machine learning and especially
recommenders brings several orders of magnitude more data than you may be used to.


On Mar 20, 2017, at 5:58 AM, Kasper Bjerke <kasper.bjerke91@gmail.com> wrote:

Thank you for a quick response. 

And thank you for clearing up the unknown/new user issue. In my total lack of imagination
I didn't think of passing an empty user parameter. 

I looked into the blacklisting, is it possible to also use this function to blacklist categories?
Say a user already selected an item of a certain category, and now doesn't want to keep getting
recommendations in the same category, would I have to pass all categories and exclude the
the ones already used? 

Initially I wanted to use the UR, but the owner did not want to use HBase and Elasticsearch
at this point. I'm fairly new to this type of technology, but maybe I will try to learn it
and try that out at a later point.

Kasper

On 20 March 2017 at 13:38, Daniel O' Shaughnessy <danieljamesdavid@gmail.com <mailto:danieljamesdavid@gmail.com>>
wrote:
Hi Kasper,

The Ecom Recommender (http://predictionio.incubator.apache.org/templates/ecommercerecommendation/quickstart/
<http://predictionio.incubator.apache.org/templates/ecommercerecommendation/quickstart/>)
actually does have a basic way to handle unknown/new users, they get recommended generally
popular items.

I think the Universal Recommender might suit your needs best though. You can find out more
about it there : http://actionml.com/docs/ur <http://actionml.com/docs/ur>

And yes, AFAIK the blacklist option will simply filter out any items you pass to it.

Regards,

Daniel. 


On Mon, 20 Mar 2017 at 10:02 Kasper Bjerke <kasper.bjerke91@gmail.com <mailto:kasper.bjerke91@gmail.com>>
wrote:
Hey guys, I'm currently working on a recommender engine for e-commerce. I chose the official
E-com Recommender. I have successfully trained and deployed it. And have been testing some
simple requests:
{ "user" : "someUserId", "num" : 5 } etc. and I get some nice responses.

Now I was wondering if there's some documentation on the following:
- Provide recommendations for users who are not logged in, like first time users who would
sign up during the purchase or just people who forgot to sign in before adding items to cart.
- And is there a way to pass the existing items in cart as a parameter to get a better recommendation?
// is this the blackList Set[ ]?

Thank you in advance,
Kasper



Mime
View raw message