mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sebastian Schelter <...@apache.org>
Subject Re: Similarity Matrix without preferences
Date Wed, 26 Sep 2012 14:23:12 GMT
The similarity computation as it is currently implemented is not
described in Mahout in Action.

We wrote a paper about the current approach which was published at this
years RecSys conference:

http://ssc.io/wp-content/uploads/2012/06/rec11-schelter.pdf

If the preferences are boolean, you simply have to use a similarity
measure that can handle such data, e.g. jaccard/tanimoto or
loglikelihood ratio.

--sebastian

On 26.09.2012 16:10, William Quinones wrote:
> Hello, I am currently writing my Bachelor thesis and am using Mahout for a
> recommendation system. I bought the book Mahout in Action and understand
> how the similarity matrix works for the distributed RecommenderJob.
> However, this explanation only shows how it works when the user has
> expressed some preferences towards the items. I would like to know how this
> works when there are no preferences (booleanData true). Are all preference
> values somehow inferred? Or treated as 1? I would also like to know what
> role the similarity metrics play in these Map/Reduce implementation.
> 
> Thank you very much in advance,
> William
> 


Mime
View raw message