You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@predictionio.apache.org by Vaghawan Ojha <va...@gmail.com> on 2017/10/23 11:37:40 UTC

Re: PIO Template for User Label Inference

Hi, you could use the labels as properties, templates like UR supports
multiple properties, you can set the properties for that user.

http://actionml.com/docs/ur_input
https://predictionio.incubator.apache.org/datacollection/eventapi/ you can
see the uses of properties in the events here.

Thanks

On Mon, Oct 23, 2017 at 4:32 PM, liuyinwei <li...@juqitech.com> wrote:

> *Hi there,*
>
>
> * Could someone know which PIO template can do customer label/profile
> inference according to the product label?*
>
>  We want to use PIO to infer the user profile and user labels according to
> the products they viewed or bought,  our customers could listen and buy
> some songs in our website, and then we can know the preference of
> customer.
>
> * here is one example:  one Customer A listened song:  *
>   * Numb -- (labels: linkin park, Rock , Pop ),
>   * New Divided -- (labels: linkin park, Rock, Pop ),
>   *  Hotel California -- (lables: The Eagles, Rock , 70's)
>
> Then, we know *Customer A should have user label (Rock) in its user
> profile.*
>
> *Is there a PIO Template which can do User Lable Inference?*
>
> *David*
>
>
>

Re: PIO Template for User Label Inference

Posted by Pat Ferrel <pa...@occamsmachete.com>.
No the current separation of concerns is:
Recommender returns recommended item ids
User’s DB keeps all item metadata, much of which has nothing to do with data the recommender needs. 
The application queries the Recommenders for item-ids which are keys into the DB that retrieve item metadata

If we were to make the Recommender the store of record for item metadata it would have to keep a a lot of things it doesn’t use to drive things like UIs or things that it can’t represent (it’s not a DB). This is generally better separated to something designed to be a general purpose store.

That said, the metadata the IS put into the recommender is actually returned for each item but stripped before returning. Fairly easy to change.


On Oct 24, 2017, at 8:43 PM, liuyinwei <li...@juqitech.com> wrote:

Thanks Vaghawan. We tried UR but it doesn't return recommended User Label/Properties, instead it only return recommended items. 
Am I miss anything here?

Here is the UR query we use to get user data: query: { "user": "U 2" ,"num":4}

Here is the UR response, however, it doesn't have properties we want, ideally, we want it return user's properites, e.g. 70s
{
    "itemScores": [
        {
            "item": "Iphone 4",
            "score": 0.11071484535932541
        },
        {
            "item": "Ipad-retina",
            "score": 0.011783885769546032
        },
        {
            "item": "Surface",
            "score": 0
        }
    ]
}

David

On 10/23/2017 19:37，Vaghawan Ojha<va...@gmail.com> <ma...@gmail.com> wrote： 
Hi, you could use the labels as properties, templates like UR supports multiple properties, you can set the properties for that user. 

http://actionml.com/docs/ur_input <http://actionml.com/docs/ur_input>
https://predictionio.incubator.apache.org/datacollection/eventapi/ <https://predictionio.incubator.apache.org/datacollection/eventapi/> you can see the uses of properties in the events here. 

Thanks

On Mon, Oct 23, 2017 at 4:32 PM, liuyinwei <liuyinwei@juqitech.com <ma...@juqitech.com>> wrote:
Hi there,

 Could someone know which PIO template can do customer label/profile inference according to the product label?

 We want to use PIO to infer the user profile and user labels according to the products they viewed or bought,  our customers could listen and buy some songs in our website, and then we can know the preference of customer. 

 here is one example:  one Customer A listened song:  
  * Numb -- (labels: linkin park, Rock , Pop ), 
  * New Divided -- (labels: linkin park, Rock, Pop ), 
  *  Hotel California -- (lables: The Eagles, Rock , 70's)

Then, we know Customer A should have user label (Rock) in its user profile.

Is there a PIO Template which can do User Lable Inference?

David

Re: PIO Template for User Label Inference

Posted by Vaghawan Ojha <va...@gmail.com>.
Hi, Yes as Pat put it clearly as he always does, recommender doesn't do
that, if you really need that maybe you wanna map the events data of the
event server with the user_id and fetch their properties/profile.

Thanks
Vaghawan

On Wed, Oct 25, 2017 at 9:28 AM, liuyinwei <li...@juqitech.com> wrote:

> Thanks Vaghawan. We tried UR but it doesn't return recommended User
> Label/Properties, instead it only return recommended items.
> Am I miss anything here?
>
> *Here is the UR query we use to get user data:* *query: { "user": "U 2"
> ,"num":4}*
>
> *Here is the UR response, however, it doesn't have properties we want,
> ideally, we want it return user's properites, e.g. 70s*
> {
>     "itemScores": [
>         {
>             "item": "Iphone 4",
>             "score": 0.11071484535932541
>         },
>         {
>             "item": "Ipad-retina",
>             "score": 0.011783885769546032
>         },
>         {
>             "item": "Surface",
>             "score": 0
>         }
>     ]
> }
>
> David
>
> On 10/23/2017 19:37，Vaghawan Ojha<va...@gmail.com>
> <va...@gmail.com> wrote：
>
> Hi, you could use the labels as properties, templates like UR supports
> multiple properties, you can set the properties for that user.
>
> http://actionml.com/docs/ur_input
> https://predictionio.incubator.apache.org/datacollection/eventapi/ you
> can see the uses of properties in the events here.
>
> Thanks
>
> On Mon, Oct 23, 2017 at 4:32 PM, liuyinwei <li...@juqitech.com> wrote:
>
>> *Hi there,*
>>
>>
>> * Could someone know which PIO template can do customer label/profile
>> inference according to the product label?*
>>
>>  We want to use PIO to infer the user profile and user labels according
>> to the products they viewed or bought,  our customers could listen and buy
>> some songs in our website, and then we can know the preference of
>> customer.
>>
>> * here is one example:  one Customer A listened song:  *
>>   * Numb -- (labels: linkin park, Rock , Pop ),
>>   * New Divided -- (labels: linkin park, Rock, Pop ),
>>   *  Hotel California -- (lables: The Eagles, Rock , 70's)
>>
>> Then, we know *Customer A should have user label (Rock) in its user
>> profile.*
>>
>> *Is there a PIO Template which can do User Lable Inference?*
>>
>> *David*
>>
>>
>>
>