mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Steven Bourke <sbou...@gmail.com>
Subject Re: Evaluating a recomender performance
Date Wed, 18 Jan 2012 14:55:12 GMT
Here is a link to a simple white paper

http://www.contentwise.tv/files/Do-Metrics-Make-Recommender-Algorithms.pdf

They have an evaluation methodologies section where they explain quite simply how to measure
your recommendation quality. 

On 18 Jan 2012, at 14:32, Raphael Bauduin wrote:

> Hi,
> 
> I'm looking to the Recomender documentation, and it looks quite
> straight-forward to use.
> I'm using data from an ecommerce website and would like to build a
> model to suggest products based on past purchases ( not on ratings ).
> I'd like to be able to compare several options, however, I found no
> mention of evaluating the performance of the recommender.
> In my case, the recommendation would be a success if the test user
> (not present in the learning data and thus used for building the
> model) actually bought (one of ) the suggested  product(s).
> 
> Does anyone have suggestions on performance measure for this kind of
> Recomender? I'm interested in any and all tips!
> 
> Thanks
> 
> Raphaƫl


Mime
View raw message