First, it appears that you are using the "dismax" query parser, not the
extended dismax ("edismax") query parser.
My hunch is that some of those fields may be non-tokenized "string" fields
in which one or more of your search keywords do appear but not as the full
string value or maybe with a different case than in the query. But when you
do a copyField from a string field to a tokenized "text" field those strings
would be broken up into individual keywords and probably lowercased. So, it
will be easier for a document to match the combined "text" field than the
source "string" fields. A fair percentage of the terms may occur in both
"text" and "string" fields, but it looks like a fair percentage may occur
only in the string fields.
Identify a specific document that is returned by the first query and not the
second. Then examine each non-text "string" field value of that document to
see if the query terms would match after text field analysis but are not
exact string matches for the string fields in which the terms do occur.
-- Jack Krupansky
-----Original Message-----
From: André Maldonado
Sent: Wednesday, June 06, 2012 9:23 AM
To: solr-user@lucene.apache.org
Subject: Re: ExtendedDisMax Question - Strange behaviour
Erick, thanks for your reply and sorry for the confusion in last e-mail.
But it is hard to explain the situation without that bunch of code.
...
