neural networks and filePro

[GCC Consulting]

> -----Original Message-----
> From: filepro-list-bounces at lists.celestial.com 
> [mailto:filepro-list-bounces at lists.celestial.com] On Behalf 
> Of Bill Vermillion
> Sent: Tuesday, November 16, 2004 6:46 PM
> To: filePro List
> Subject: Re: neural networks and filePro
> 
> On Tue, Nov 16 14:52 , Walter Vaughan moved his mouse, 
> rebooted for the change to take effect, and then said:" 
> 
> > John Esak wrote:
> 
> > >>Has anyone fed their filePro derived order - line item data at a 
> > >>neural network to see any associations?
> 
> > >Do you mean _other_ than my own neural net?
> 
> > No. I'm really serious.
> 
> > Do customers order in patterns? Does the purchase of one 
> item indicate 
> > the purchase of something later?
> 
> I'm imressed with the matching which is done by Amazon on 
> recommnedations - and my tastes are so widely scattered.
> 
> It inspired me to check what the recommendations are at this 
> moment and I see Koyaanisqatsi / Powaqqatsi in the recommend list.
> I own those and it also listed Kurosawa's Dreams - which I also own
> - and those aren't what you'd call mainstream - so they 
> definitely are doing something right.
> 
> What I like about Amazon is they have a 'why was I 
> recommended this' - and one of the reasons was Kieslowski's 
> Three Colors trilogy.
> 
> About a year or so ago Sears was expanding their data base of 
> customer purchases to be able to analyze purchases by a 
> customer and make recommendations.  One example was if 
> something was bought in kids swimming suits, and luggage 
> bought by adults, they might conclude a trip was in the 
> offing and make offers for those.
> 
> They expected their database to go from 7 Terabytes to over 
> 70 terabytes for that analysis ability.
> 
> Trends and tastes based on past purchases are proving to be 
> valid for at least some places.  
> 
> ...
> 
> > Has anyone applied data they generated with filePro to a neural 
> > network to learn behavior?
> 
> But don't you have to analyze and give some criteria to each 
> thing bought to be able to make predictions?  To me that 
> would be the hardest part - doing the analysis of data - and 
> not the storing/collecting.

Super markets and other brick & Mortar retailers are doing this with their
affinity programs.

However, there is a problem with all of this.

Case 1: Amazon

I purchased THE Atkins Book for my mother in law.  I am 6.1 and weigh 175
pounds.  I kept receiving e-mail, snail mail and phone calls about diets.

Case 2: Foodtown Supermarket

Our vet suggested that we put our older cat on a carb diet.  So we tried various
infants pasta based foods. The cat eat these foods for about a week.  However,
we kept getting offers each time we checked out for children's food.  This
happened for almost a year and 1/2. No matter what I did at the store level, I
couldn't stop these 'suggestions" for children's food.

Finally in desperation, I sent a letter the CEO of Foodtown.  I told him what
was happening and then said that he may be loosing business because of this.

First of all it may have been grandparents purchasing the item when a grandchild
stayed with them.  Once the kid left, they didn't need the food.

I think the next reason may have gotten to them; I said, you may be loosing a
mother whose child past away and this constant reminder of her lose would cause
her to stop shopping in his store.

The last suggestion I made was to a date check to see if the person has stopped
purchasing this category of merchandise for some fixed interval.

Within 2 weeks of my letter, the toddler food items stopped coming up.

Data mining can be beneficial, but it has to have some "smarts" to read changes.

Richard Kreiss
GCC Consulting