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 2017/04/24 08:10:15 UTC

Knowing the probability of purchase

Hi,

I was following a research paper regarding the probability of a user buying
a particular item recommended by the recommendation system. It's here, if
you want to checkout as well
http://www.kecl.ntt.co.jp/as/members/iwata/doctor.pdf

I was wondering if there is a way or anybody has done with the current
templates of PIO, the calculation of the probability of a user buying an
item.

I think this should be possible with current templates as well, I am just
wondering, if anybody could provide me a brief way to do that, or any
documentation of the algorithms that could be used.


Thanks

Re: Knowing the probability of purchase

Posted by Vaghawan Ojha <va...@gmail.com>.

Hi Pat,

Yes, you're correct, it's quite vain to try to weigh the recommendation
with the probability. And also probability is quite plain in terms of
complex colloborative filtering. It's no good to use probability matrics
but in my case I  was also experimenting the offline sales data, which
doesn't need the same sorts of result as we may need for E-commerce.

Anyways, I used spark sql to do some analytics and calculate some plain
probability in terms of physcial stores and sales. You have been very
helpful to me. Thank you very much for your time and help.

Thanks

On Wed, Apr 26, 2017 at 11:16 PM, Pat Ferrel <pa...@occamsmachete.com> wrote:

> There is a subtle but very important point here.
>
> This does not give you the probability of a user buying a recommendation
> it gives you the probability that a user will buy a product. You will see
> that a user buys a very small % products at Amazon, so if the technique is
> extremely accurate it will tell you the probability is 0 for the vast
> majority of products.
>
> A recommender works quite differently, getting a user to convert on things
> that can’t be predicted from only their history. This is the "collaborative
> filtering” part. The recommender looks at the behavior of other people to
> recommend, therefor it is not trying to answer the question of what an
> individual is going to do but what they might do if given the choice. In
> this sense it is part of your discovery tools including search and browsing.
>
> You might combine the 2 by getting recommendations, then asking this other
> technique for the probability a user will buy any of the recommendations
> but what would you do with the information? They may seem like that same
> thing but one will lead to conversion lift and (if the other operates
> perfectly) the other will not.
>
>
> On Apr 26, 2017, at 5:04 AM, Vaghawan Ojha <va...@gmail.com> wrote:
>
> Hi Pat,
>
> Yes, I understood  that, but it was sorts of curiosity to know what would
> be the precise probability for a user to buy such recommended product. I do
> understand, "The job of a recommender is not to predict what you *will*
> buy but rather what you would like to buy"
>
> Thank you for your prompt reply.
>
> Thanks
>
> On Wed, Apr 26, 2017 at 12:28 AM, Pat Ferrel <pa...@occamsmachete.com>
> wrote:
>
>> I read this dissertation and came away wondering why it was important.
>> The job of a recommender is not to predict what you *will* buy but rather
>> what you would like to buy if you knew about it—in other words it
>> determines your taste or preferences and finds item that match. This tends
>> to increase conversions (sales for E-Commerce). A predictor may only
>> predict the inevitable and lead to 0 lift in conversions.
>>
>>
>> On Apr 24, 2017, at 1:10 AM, Vaghawan Ojha <va...@gmail.com> wrote:
>>
>> Hi,
>>
>> I was following a research paper regarding the probability of a user
>> buying a particular item recommended by the recommendation system. It's
>> here, if you want to checkout as well http://www.kecl.ntt.co.jp
>> /as/members/iwata/doctor.pdf
>>
>> I was wondering if there is a way or anybody has done with the current
>> templates of PIO, the calculation of the probability of a user buying an
>> item.
>>
>> I think this should be possible with current templates as well, I am just
>> wondering, if anybody could provide me a brief way to do that, or any
>> documentation of the algorithms that could be used.
>>
>>
>> Thanks
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "actionml-user" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to actionml-user+unsubscribe@googlegroups.com.
>> To post to this group, send email to actionml-user@googlegroups.com.
>> To view this discussion on the web visit https://groups.google.com/d/ms
>> gid/actionml-user/CA%2B69RXZnXGY_wMbU55fE_AW7Ur8YRyBkBJ%2BWc
>> QXLKjRtPGi6tg%40mail.gmail.com
>> <https://groups.google.com/d/msgid/actionml-user/CA%2B69RXZnXGY_wMbU55fE_AW7Ur8YRyBkBJ%2BWcQXLKjRtPGi6tg%40mail.gmail.com?utm_medium=email&utm_source=footer>
>> .
>> For more options, visit https://groups.google.com/d/optout.
>>
>>
>
> --
> You received this message because you are subscribed to the Google Groups
> "actionml-user" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to actionml-user+unsubscribe@googlegroups.com.
> To post to this group, send email to actionml-user@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/
> msgid/actionml-user/CA%2B69RXYZ0FhXP%2BqWSJGSOW%2B-
> vVY2d7CVZhS_smSNgPpzSON%3DBg%40mail.gmail.com
> <https://groups.google.com/d/msgid/actionml-user/CA%2B69RXYZ0FhXP%2BqWSJGSOW%2B-vVY2d7CVZhS_smSNgPpzSON%3DBg%40mail.gmail.com?utm_medium=email&utm_source=footer>
> .
> For more options, visit https://groups.google.com/d/optout.
>
>

Re: Knowing the probability of purchase

Posted by Pat Ferrel <pa...@occamsmachete.com>.

There is a subtle but very important point here.

This does not give you the probability of a user buying a recommendation it gives you the probability that a user will buy a product. You will see that a user buys a very small % products at Amazon, so if the technique is extremely accurate it will tell you the probability is 0 for the vast majority of products.

A recommender works quite differently, getting a user to convert on things that can’t be predicted from only their history. This is the "collaborative filtering” part. The recommender looks at the behavior of other people to recommend, therefor it is not trying to answer the question of what an individual is going to do but what they might do if given the choice. In this sense it is part of your discovery tools including search and browsing.

You might combine the 2 by getting recommendations, then asking this other technique for the probability a user will buy any of the recommendations but what would you do with the information? They may seem like that same thing but one will lead to conversion lift and (if the other operates perfectly) the other will not.

On Apr 26, 2017, at 5:04 AM, Vaghawan Ojha <va...@gmail.com> wrote:

Hi Pat,

Yes, I understood that, but it was sorts of curiosity to know what would be the precise probability for a user to buy such recommended product. I do understand, "The job of a recommender is not to predict what you *will* buy but rather what you would like to buy"

Thank you for your prompt reply.

Thanks

On Wed, Apr 26, 2017 at 12:28 AM, Pat Ferrel <pat@occamsmachete.com <ma...@occamsmachete.com>> wrote:
I read this dissertation and came away wondering why it was important. The job of a recommender is not to predict what you *will* buy but rather what you would like to buy if you knew about it—in other words it determines your taste or preferences and finds item that match. This tends to increase conversions (sales for E-Commerce). A predictor may only predict the inevitable and lead to 0 lift in conversions.

On Apr 24, 2017, at 1:10 AM, Vaghawan Ojha <vaghawan781@gmail.com <ma...@gmail.com>> wrote:

Hi,

I was following a research paper regarding the probability of a user buying a particular item recommended by the recommendation system. It's here, if you want to checkout as well http://www.kecl.ntt.co.jp/as/members/iwata/doctor.pdf <http://www.kecl.ntt.co.jp/as/members/iwata/doctor.pdf>

I was wondering if there is a way or anybody has done with the current templates of PIO, the calculation of the probability of a user buying an item.

I think this should be possible with current templates as well, I am just wondering, if anybody could provide me a brief way to do that, or any documentation of the algorithms that could be used.

Thanks

--
You received this message because you are subscribed to the Google Groups "actionml-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to actionml-user+unsubscribe@googlegroups.com <ma...@googlegroups.com>.
To post to this group, send email to actionml-user@googlegroups.com <ma...@googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/CA%2B69RXZnXGY_wMbU55fE_AW7Ur8YRyBkBJ%2BWcQXLKjRtPGi6tg%40mail.gmail.com <https://groups.google.com/d/msgid/actionml-user/CA%2B69RXZnXGY_wMbU55fE_AW7Ur8YRyBkBJ%2BWcQXLKjRtPGi6tg%40mail.gmail.com?utm_medium=email&utm_source=footer>.
For more options, visit https://groups.google.com/d/optout <https://groups.google.com/d/optout>.

--
You received this message because you are subscribed to the Google Groups "actionml-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to actionml-user+unsubscribe@googlegroups.com <ma...@googlegroups.com>.
To post to this group, send email to actionml-user@googlegroups.com <ma...@googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/CA%2B69RXYZ0FhXP%2BqWSJGSOW%2B-vVY2d7CVZhS_smSNgPpzSON%3DBg%40mail.gmail.com <https://groups.google.com/d/msgid/actionml-user/CA%2B69RXYZ0FhXP%2BqWSJGSOW%2B-vVY2d7CVZhS_smSNgPpzSON%3DBg%40mail.gmail.com?utm_medium=email&utm_source=footer>.
For more options, visit https://groups.google.com/d/optout <https://groups.google.com/d/optout>.

Re: Knowing the probability of purchase

Posted by Vaghawan Ojha <va...@gmail.com>.

Hi Pat,

Yes, I understood  that, but it was sorts of curiosity to know what would
be the precise probability for a user to buy such recommended product. I do
understand, "The job of a recommender is not to predict what you *will* buy
but rather what you would like to buy"

Thank you for your prompt reply.

Thanks

On Wed, Apr 26, 2017 at 12:28 AM, Pat Ferrel <pa...@occamsmachete.com> wrote:

> I read this dissertation and came away wondering why it was important. The
> job of a recommender is not to predict what you *will* buy but rather what
> you would like to buy if you knew about it—in other words it determines
> your taste or preferences and finds item that match. This tends to increase
> conversions (sales for E-Commerce). A predictor may only predict the
> inevitable and lead to 0 lift in conversions.
>
>
> On Apr 24, 2017, at 1:10 AM, Vaghawan Ojha <va...@gmail.com> wrote:
>
> Hi,
>
> I was following a research paper regarding the probability of a user
> buying a particular item recommended by the recommendation system. It's
> here, if you want to checkout as well http://www.kecl.ntt.co.
> jp/as/members/iwata/doctor.pdf
>
> I was wondering if there is a way or anybody has done with the current
> templates of PIO, the calculation of the probability of a user buying an
> item.
>
> I think this should be possible with current templates as well, I am just
> wondering, if anybody could provide me a brief way to do that, or any
> documentation of the algorithms that could be used.
>
>
> Thanks
>
> --
> You received this message because you are subscribed to the Google Groups
> "actionml-user" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to actionml-user+unsubscribe@googlegroups.com.
> To post to this group, send email to actionml-user@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/
> msgid/actionml-user/CA%2B69RXZnXGY_wMbU55fE_AW7Ur8YRyBkBJ%
> 2BWcQXLKjRtPGi6tg%40mail.gmail.com
> <https://groups.google.com/d/msgid/actionml-user/CA%2B69RXZnXGY_wMbU55fE_AW7Ur8YRyBkBJ%2BWcQXLKjRtPGi6tg%40mail.gmail.com?utm_medium=email&utm_source=footer>
> .
> For more options, visit https://groups.google.com/d/optout.
>
>

Re: Knowing the probability of purchase

Posted by Pat Ferrel <pa...@occamsmachete.com>.

I read this dissertation and came away wondering why it was important. The job of a recommender is not to predict what you *will* buy but rather what you would like to buy if you knew about it—in other words it determines your taste or preferences and finds item that match. This tends to increase conversions (sales for E-Commerce). A predictor may only predict the inevitable and lead to 0 lift in conversions.

On Apr 24, 2017, at 1:10 AM, Vaghawan Ojha <va...@gmail.com> wrote:

Hi,

I was wondering if there is a way or anybody has done with the current templates of PIO, the calculation of the probability of a user buying an item.

I think this should be possible with current templates as well, I am just wondering, if anybody could provide me a brief way to do that, or any documentation of the algorithms that could be used.

Thanks