lucene-general mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Dunning <ted.dunn...@gmail.com>
Subject Re: Score Normalization in Lucene 3.02
Date Tue, 14 Dec 2010 16:48:24 GMT
Scores are not comparable across different queries.  There are lots of
reasons, but the real, deep, dark
reason is that the decision was taken very early on in Lucene development to
not try to make scores
comparable across queries, or even combinable.  The result is as you see.
 You can't easily combine
these scores.

On Tue, Dec 14, 2010 at 6:48 AM, vermansi <vermansi@gmail.com> wrote:

>
> Hello
> I had a doubt in scoring of documents in lucene.
> If there is an OR query like "w1 OR w2 OR w3".
> One way is to fire the query and get the results. The scores that come with
> the documents are not normalized. right ? .. as they do exceed 1.
>
> The second way is to fire w1 and then w2 and then w3 and combine the
> results. My problem as well as the doubt is the results of both these
> approaches differ greatly. The documents dont match if we compare the
> combined scores (using each word at a time as a query) and original scores
> (using w1 OR w2 OR w3).
>
> since there seems to be no normalization of scores in 3.02. why is this
> difference occuring . Is it due to createWeight function in Searcher.java.
> Im sorry if i have missed on how the documents are scored in 3.02 though i
> checked the previos posts, they all talk about Hits.java which is not used
> anymore.
>
> So my question is --
> is the difference due to normalization. If yes then how can i fix it. (my
> query is a fielded query ie  --> f1:w1 OR f2:w2  but none of the fields has
> any weights)
> I really need to use the second approach as my queries are very big.
>
> Regards
> Manisha
> --
> View this message in context:
> http://lucene.472066.n3.nabble.com/Score-Normalization-in-Lucene-3-02-tp2085804p2085804.html
> Sent from the Lucene - General mailing list archive at Nabble.com.
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message