The Core

Why We Are Here => Web Development => Topic started by: Chunkford on April 24, 2014, 07:29:00 PM

Title: Recommender sytems...
Post by: Chunkford on April 24, 2014, 07:29:00 PM
Anyone had one developed or know someone that has?
Been reading up on it and makes for good bed time reading, but Interested in what sort of $$$ I'm looking at so I don't keel over when I start looking.
Title: Re: Recommender sytems...
Post by: ergophobe on April 24, 2014, 10:30:53 PM
You mean like Amazon's "also bought" feature?
Title: Re: Recommender sytems...
Post by: Chunkford on April 25, 2014, 11:24:54 AM
Yea, but not for customer use. It's for behind the scenes so custom orders can be created.
Looking at the aglos is scary enough, but I see there are a few frameworks to use as a starting point.
The question is purely so I know what sort of investment I need to be aiming for, so when I hit that level I can start implementing it.
Title: Re: Recommender sytems...
Post by: JasonD on April 25, 2014, 12:37:16 PM
As it so happens ..... :)

Check out - http://www.mortardata.com/ for a larger data set and a simple(ish) way of getting the data out using a SAAS(ish) app.

If the data sets are smaller then may I suggest you look into market basket analysis, the simplest form of which is likely the apriori algorithm. There are quite a few libraries and codebases to help you out on that curve.

https://github.com/h-collector/Apriori
Title: Re: Recommender sytems...
Post by: Adam C on April 30, 2014, 01:43:18 PM
I'm not sure whether this will be useful for you, but I found it an interesting read:

http://patft.uspto.gov/netacgi/nph-Parser?Sect2=PTO1&Sect2=HITOFF&p=1&u=/netahtml/PTO/search-bool.html&r=1&f=G&l=50&d=PALL&RefSrch=yes&Query=PN/8447747

QuoteSystem for generating behavior-based associations for multiple domain-specific applications

Abstract

Systems and processes are described for generating a generic relationship dataset that can be adapted for use by multiple domain-specific applications. Such applications can include recommendation applications, search applications, optimization applications, and/or other applications. These systems and processes can analyze some or all actions of users with respect to a network application, such as a web application. Data representing user actions occurring in a time window can be paired together in the relationship dataset, together with one or more metrics related to the user actions. These metrics can represent general measurements that in certain embodiments are not customized for an individual application. As a result, separate domain-specific applications can independently analyze the relationship data according to different criteria without maintaining separate sets of the relationship data.
Title: Re: Recommender sytems...
Post by: Chunkford on April 30, 2014, 04:34:45 PM
oooooo Pandora's Box has just been passed over to me.
Cheers, will have a read guys :)
Title: Re: Recommender sytems...
Post by: jetboy on April 30, 2014, 07:40:35 PM
http://th3core.com/talk/web-development/deconstructing-recommender-systems/