Hi all.
We have 1..N indexes for each time someone adds some data. Each time
they can choose different tokenisation settings. Because of this, each
text index has its own query parser instance. Because each query
parser could generate a different Query (though I guess whether they
do or not is another matter!) we can't just use MultiReader and a
single IndexSearcher.
What we have been doing is performing the search against each index
and then merging the results by score. I remember from a long time ago
that this was not something you're supposed to do.
Reading information I can find today, it sounds like the main problem
was Hits doing score normalisation. We cloned Hits a while back and
inherited its score normalisation code as well.
So my questions:
- does normalising the scores do anything other than give people the
illusion that scores are from 0.0 ~ 1.0?
- if I remove the score normalisation from our Hits-like class, is
it now totally OK to merge the lists of results?
(And, I guess, if not, is there some trick I can employ to line
the scores for the results up so that I can?)
TX
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org
For additional commands, e-mail: java-user-h...@lucene.apache.org
