[ 
https://issues.apache.org/jira/browse/LUCENE-5489?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13939069#comment-13939069
 ] 

Simon Willnauer commented on LUCENE-5489:
-----------------------------------------

it's awesome that we are adding this. I have almost the same code on a high 
level in Elasticsearch and it works just great. Ideally I'd love to drop the 
code from ES and replace it with the lucene one but I think we need to iterate 
more on this. From what I can see the biggest issues are :

 * the way how scores are combined is hardcoded (we just multiply)
 * we only have one way to rescore ie. we use another query by default which 
would be nice if the interface would allow us to use other ways of rescorers.
 * most of the goodies are hidden in a static method I think we should add a 
nice interface / abstract class 
 * it would be awesome if the interface could provide a way to get an 
Explanation like other queries...

what I think would make sense is something like this:

{code}

public class AbstractRescorer implements Rescore {

  @Override
  public TopDocs rescore(IndexSearcher searcher, TopDocs topDocs, int topN) {
     // do what is done int he static method
  }

  protected abstract float combine(float originalScore, float resocredScore);
  
  @Override 
  public Explanation explain(IndexSearcher searcher, Explanation sourceExplain, 
int docId) {
    // impl explain
  }

}
{code}

I hope this makes sense?




> Add query rescoring API
> -----------------------
>
>                 Key: LUCENE-5489
>                 URL: https://issues.apache.org/jira/browse/LUCENE-5489
>             Project: Lucene - Core
>          Issue Type: Improvement
>            Reporter: Michael McCandless
>            Assignee: Michael McCandless
>             Fix For: 4.8, 5.0
>
>         Attachments: LUCENE-5489.patch, LUCENE-5489.patch
>
>
> When costly scoring factors are used during searching, a common
> approach is to do a cheaper / basic query first, collect the top few
> hundred hits, and then rescore those hits using the more costly
> query.
> It's not clear/simple to do this with Lucene today; I think we should
> make it easier.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to