mahout-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tolga Oral (JIRA)" <j...@apache.org>
Subject [jira] Commented: (MAHOUT-432) Can add one parameter --itemsFile for org.apache.mahout.cf.taste.hadoop.item.RecommenderJob
Date Thu, 01 Jul 2010 18:19:51 GMT

    [ https://issues.apache.org/jira/browse/MAHOUT-432?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12884371#action_12884371
] 

Tolga Oral commented on MAHOUT-432:
-----------------------------------

Hi Sean;

This feature is available on the non-distributed algo's vi IDRescorer, its pretty powerfull
to be skipped over for the distributed version.

I havent had a chance to look at the implementation of distributed recommender, however reading
the Mahout In Action, it seems like another map/reduce task might do the job.

Conceptually additional step would generate an identity matrix where all items except the
ones you are able to recommend will have 0 at their index. Also same matrix can be used to
boost recommendations like the rescore() method when value of the identity matrix is not 1
but a different scalar multiplier.

What do you think Sean, would this work? We can look at how to implement it if sounds like
a possible solution to you?

> Can add one parameter --itemsFile for org.apache.mahout.cf.taste.hadoop.item.RecommenderJob

> --------------------------------------------------------------------------------------------
>
>                 Key: MAHOUT-432
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-432
>             Project: Mahout
>          Issue Type: New Feature
>          Components: Collaborative Filtering
>    Affects Versions: 0.4
>            Reporter: Hui Wen Han
>            Assignee: Sean Owen
>            Priority: Minor
>             Fix For: 0.4
>
>
> Can add one parameter --itemsFile for org.apache.mahout.cf.taste.hadoop.item.RecommenderJob
?
> --itemsFile some likes the the --usersFile.
> in some case, we want to calculate similarity using all preferences data ,
> but some item in  preferences  are old ,not available,behind the times or retired ,
> we want that those data can not recommend to users, can only recommend user with new
or valid items.
> so we can add on paramete  --itemsFile for RecommenderJob, itemsFile includes all the
new valid items.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message