[vwnc] Algorithms for future purchase predictions for an ordering system?

Hi all..

One of the areas I'll be working on this summer for a Seaside based
web-app has to do with making suggestions for ordering products based on
past buying habits (taking into account perhaps the past 4-5 years).  The
current people-driven system (no computers helping currently) works on a
demand schedule -- if people ask for product X then that product is
ordered -- causing people to wait for delivery.  What I want to do is look
back in time perhaps a month in advance (say for the Christmas period) and
have the system make suggestions for products to buy that have been
popular for the past several years.  That way when people put in their
orders, the product is sitting and waiting for them instead of the other
way around.

To that end, I could certainly generate some sort of algorithm to look at
purchase history for a selected time period and make suggestions..
However, I figured it might be best to ask if anyone knows of an algorithm
that tries to solve this equation from the get-go... Anyway, if no such
algorithm exists then I'll brute-force it and write my own but thought I'd
ask first.. Thx!



_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

Rick,

If you do not find a full package for purchase-pattern analysis, I do
still suggest that you consider the excellent histogramming capabilities
provided in the Numerical Methods package in the public repository.  
This implements the algorithms from the book "Object-oriented
implementation of numerical methods" by Didier H. Besset, ISBN
1-55860-679-3.   The book is worth having on hand too.

This application area can be addressed by producing histograms for
periods of interest in the past, then detecting whether the histograms
have high enough peaks and/or enough area (volume) to warrant increasing
planned orders beyond the routine levels.

Gregory Bourassa

Rick Flower wrote:
_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

Gregory,

Thanks for the great suggestions!  I figured I was not the only one to
have done this.. I'm hoping that by implemented this along with a well
managed inventory system, that I can ensure a stress free purchasing
experience for the users of the system -- at least much better than how it
runs now.

On Tue, June 24, 2008 9:34 am, Gregory Bourassa wrote:

_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

I find that the Book "artificial Intelligence A modern approach 2nd
edition" by Stuart Russell and Peter Norvig  can offer some good
approaches take. Page 344 section 10.5 deals with the problem of finding
relevant offers. Which when you think about it is very similar to your
problem you are solving. So It would get you on the right track for you
solution I would think.

You could write rules for extracting the person likes and interests from
what they buy and correlate them to the product categories and perform
another analysis to take the top ten picks etc.

That section and the following one 10.6 reasoning systems for categories
could be directly applied.

HTH,


Sean Glazier
Cincom Systems

Lead Engineer

-----Original Message-----
From: [hidden email] [mailto:[hidden email]] On
Behalf Of Rick Flower
Sent: Tuesday, June 24, 2008 9:21 AM
To: [hidden email]
Subject: [vwnc] Algorithms for future purchase predictions for an
orderingsystem?

Hi all..

One of the areas I'll be working on this summer for a Seaside based
web-app has to do with making suggestions for ordering products based on
past buying habits (taking into account perhaps the past 4-5 years).
The
current people-driven system (no computers helping currently) works on a
demand schedule -- if people ask for product X then that product is
ordered -- causing people to wait for delivery.  What I want to do is
look
back in time perhaps a month in advance (say for the Christmas period)
and
have the system make suggestions for products to buy that have been
popular for the past several years.  That way when people put in their
orders, the product is sitting and waiting for them instead of the other
way around.

To that end, I could certainly generate some sort of algorithm to look
at
purchase history for a selected time period and make suggestions..
However, I figured it might be best to ask if anyone knows of an
algorithm
that tries to solve this equation from the get-go... Anyway, if no such
algorithm exists then I'll brute-force it and write my own but thought
I'd
ask first.. Thx!



_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

On Tue, 24 Jun 2008 16:51:19 -0400
"Glazier, Sean" <[hidden email]> wrote:

> You could write rules for extracting the person likes and interests from
> what they buy and correlate them to the product categories and perform
> another analysis to take the top ten picks etc.

Or just try some simple statistics (basically "yesterday's weather") to
provide estimates. The applicability of this approach depends heavily
on the lifetime of your store items, though.

If the articles are "long-lived" (as you indicated in your OP), you
could just collect the ordered amounts per year and extrapolate from
that. This approach works very well with one of my clients, where
a human is using these indicators for managing the stock and haggling
with suppliers.

 s.
_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

Stefan,

In my case, these products are not anything that expires (not food items)
so your method could work there as well -- in some cases simpler is
better.. I've got recorded data archived (SQL database dumps) going back
approx. 3-4 years for an older version of the code I wrote in PHP.. I'll
have to find a way to move the data into my Glorpified database and then
run some stats on the data and see what is spat out.. Thanks everyone!

On Tue, June 24, 2008 3:10 pm, Stefan Schmiedl wrote:

_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

I suggested the book because it covers in detail internet agents and it
is a good read. I recommend it.

Sean

-----Original Message-----
From: [hidden email] [mailto:[hidden email]] On
Behalf Of Rick Flower
Sent: Tuesday, June 24, 2008 3:38 PM
To: Stefan Schmiedl
Cc: [hidden email]
Subject: Re: [vwnc] Algorithms for future purchase predictions for an
orderingsystem?

Stefan,

In my case, these products are not anything that expires (not food
items)
so your method could work there as well -- in some cases simpler is
better.. I've got recorded data archived (SQL database dumps) going back
approx. 3-4 years for an older version of the code I wrote in PHP.. I'll
have to find a way to move the data into my Glorpified database and then
run some stats on the data and see what is spat out.. Thanks everyone!

On Tue, June 24, 2008 3:10 pm, Stefan Schmiedl wrote:
> On Tue, 24 Jun 2008 16:51:19 -0400
> "Glazier, Sean" <[hidden email]> wrote:
>
>> You could write rules for extracting the person likes and interests
from
>> what they buy and correlate them to the product categories and
perform
>> another analysis to take the top ten picks etc.
>
> Or just try some simple statistics (basically "yesterday's weather")
to

_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

I'll plan on taking a look at all of the sources mentioned and see what I
can apply to the problem.. Thanks again to everyone!

On Tue, June 24, 2008 4:06 pm, Glazier, Sean wrote:

_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

Personally, I'd look at how to model the purchase events inside a finite
state model that records the probability of each event in a limited
context of past events.  In other words, I'd try compressing the
sequence of events, and then suggesting what the compression model says
it's more likely to happen.

Just my 2 cents,
Andres.

-----Original Message-----
From: [hidden email] [mailto:[hidden email]] On
Behalf Of Rick Flower
Sent: Tuesday, June 24, 2008 9:21 AM
To: [hidden email]
Subject: [vwnc] Algorithms for future purchase predictions for an
orderingsystem?

Hi all..

One of the areas I'll be working on this summer for a Seaside based
web-app has to do with making suggestions for ordering products based on
past buying habits (taking into account perhaps the past 4-5 years).
The current people-driven system (no computers helping currently) works
on a demand schedule -- if people ask for product X then that product is
ordered -- causing people to wait for delivery.  What I want to do is
look back in time perhaps a month in advance (say for the Christmas
period) and have the system make suggestions for products to buy that
have been popular for the past several years.  That way when people put
in their orders, the product is sitting and waiting for them instead of
the other way around.

To that end, I could certainly generate some sort of algorithm to look
at purchase history for a selected time period and make suggestions..
However, I figured it might be best to ask if anyone knows of an
algorithm that tries to solve this equation from the get-go... Anyway,
if no such algorithm exists then I'll brute-force it and write my own
but thought I'd ask first.. Thx!



_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc