The XML saga takes a turn...

My client is looking into the algorithm that some web sites use to rank
their recommended products, and would like to do the process in Universe.

Has anybody used this Apriori algorithm in Pick? I'm trying to rank products
according to category, so that when a person clicks on a pink beedy.widget,
he'll get other beedys, other widgets, and other pink things, ranked with a
confidence level using this algorithm.

I know it's available as a webservice, but do you think bringing it inside
would improve response times?

I'm thinking this may not be a widely known subject, but I've been surprised
before (and lately!)

Thanks,

Ed

On May 7, 8:53*am, "Ed Sheehan" <NOedsS... (AT) xmission (DOT) com> wrote:
I looked into this breifly when deciding whether to assign a student a
data mining project that was directly relevant to my software. I
decided against it due to the complexity of the algorithm and that I
preferred he looked at bigger variety of data mining techniques. I
thought when I looked into it that there were new algorithms, so you
might check if Apriori is still the algorithm of choice before you
would try to implement it.

My not-really-well-read and no-experience guess on what a good
strategy would be is to use a web server and just make sure it is done
asynch, so users do not wait on response times. If you do get a
strategy on this, I would be interested in knowing what you come up
with. --dawn

I've since done some looking, and Apriori seems to be pretty popular out
there. I think I'd start by creating a set of xrefs with the supports and
confidence levels stored, much how an index would appear. This might be done
daily for the day's previous tramsactions, and accumulated for some
meaningful period of time, say 30 days.

Then when someone clicks on a product, I should be able to hash directly
into a record which would have the products also purchased, and also
similary ranked products (clicked a keyboard, show a mouse too). I'm hoping
I can do most of the heavy lifting offline, so the recommendations can be
retrieved quickly. Time is much more important than space here.

I'll take your suggestion and look for other methods as well. Thanks Dawn.

Ed

"dawn" <dawnwolthuis (AT) gmail (DOT) com> wrote

On May 7, 8:53 am, "Ed Sheehan" <NOedsS... (AT) xmission (DOT) com> wrote:
I looked into this breifly when deciding whether to assign a student a
data mining project that was directly relevant to my software. I
decided against it due to the complexity of the algorithm and that I
preferred he looked at bigger variety of data mining techniques. I
thought when I looked into it that there were new algorithms, so you
might check if Apriori is still the algorithm of choice before you
would try to implement it.

My not-really-well-read and no-experience guess on what a good
strategy would be is to use a web server and just make sure it is done
asynch, so users do not wait on response times. If you do get a
strategy on this, I would be interested in knowing what you come up
with. --dawn

On May 9, 10:07*am, "Ed Sheehan" <NOedsS... (AT) xmission (DOT) com> wrote:
I did a quick search to see what I might have been thinking and I see
lots of apriori out there too, so looks like a safe bet to use it.

makes sense not to do this on the fly

Makes sense to me. I'm not looking forward to it, but I'll have to do
this at some point too, I suspect, so good luck with it. --dawn