[
https://issues.apache.org/jira/browse/MAHOUT-553?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sebastian Schelter updated MAHOUT-553:
--------------------------------------
Description: When using a weighted sum for preference estimation on boolean
data, the predicted preferences can only be 1 or NaN which is mathematically
correct but not very useful for ranking them. The distributed recommender
should therefore adapt the behavior of GenericBooleanPrefItemBasedRecommender
in that case: use the sums of similarities to rank the recommended items.
(was: When use a weighted some for preference estimation on boolean data the
predicted preferences can only be 1 or NaN which is mathematically correct but
not very useful for ranking them. The distributed recommender should adapt the
behavior of GenericBooleanPrefItemBasedRecommender in that case: use the sums
of similarities to rank the recommended items.)
> Unify ranking of boolean recommendations in distributed and non-distributed
> recommenders
> ----------------------------------------------------------------------------------------
>
> Key: MAHOUT-553
> URL: https://issues.apache.org/jira/browse/MAHOUT-553
> Project: Mahout
> Issue Type: Improvement
> Components: Collaborative Filtering
> Affects Versions: 0.5
> Reporter: Sebastian Schelter
>
> When using a weighted sum for preference estimation on boolean data, the
> predicted preferences can only be 1 or NaN which is mathematically correct
> but not very useful for ranking them. The distributed recommender should
> therefore adapt the behavior of GenericBooleanPrefItemBasedRecommender in
> that case: use the sums of similarities to rank the recommended items.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
