Mailing-List: contact java-user-help@lucene.apache.org; run by ezmlm
Precedence: bulk
Reply-To: java-user@lucene.apache.org
Received-SPF: pass (athena.apache.org: local policy)
Date: Tue, 2 Dec 2008 14:42:05 -0800 (PST)
From: Chris Hostetter <hossman_lucene@fucit.org>
To: java-user@lucene.apache.org
Subject: Re: Query time document group boosting
In-Reply-To: <DDF845E9-13FC-4AF0-8B76-B6AE4C2C5DC1@gmail.com>
Message-ID: <Pine.LNX.4.62.0812021429460.27864@radix.cryptio.net>
References: <1227714863.25256.38.camel@pc286>
 <B7D0017E-C6A1-4FB8-82A4-4509806062B6@gmail.com>
 <1227777358.25256.70.camel@pc286>
 <DDF845E9-13FC-4AF0-8B76-B6AE4C2C5DC1@gmail.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII


: > "foo AND (
: >  groupboost_A:dummy^10 OR
: >  groupboost_B:dummy OR
: >  groupboost_C:dummy^0.1 OR
: >  ...
: >  groupboost_Z:dummy
: > )"
: > 
: > With that query, it seems that only documents matching foo will result
: > in a hit and be scored?
: 
: Someone else than me needs to answer this. I know there is no optimization of
: boolean clauses, that is why I'm saying this: it is possible that the boolean
: query weight actually will be visiting all the inner clauses even though "foo"
: was not matched, i.e. all documents in the index are visited but might not all
: be scored.

No, skipTo in the Scorer (ConjunctionScorer i believe) should ensure that 
none of the groupboost clauses will be visited for any doc that doesn't 
already match foo.

i believe the pathological case you are talking about would be something 
like this...

   (docnum_is_odd AND foo) AND (docnum_is_even AND bar)

assuming every doc matches either docnum_is_even or docnum_is_odd, but no 
doc matches both, then this query will match no documents, but every 
document will be visted by the scorer.

: A cosmetic remark, I would personally choose a single field for the boosts and
: then one token per source. (groupboost:A^10 groupboost:B^1 groupboost:C^0.1).

that's a key improvement, as it helps keep the number of unique fields 
down, even if the number of sources grows without bounds.  make sure you 
omitNorms on your groupboost field, and when buiding your various boolean 
queries, consider disabling the coord (check the docs to understand why 
that might make sense)

: > > I think you are looking for CustomScoreQuery.

not neccessarily ... CustomScoreQueries make a lot of sense when you want 
the score to be a function of the field value, but for simple "exists or 
not" fields a BooleanQuery works just as well.

(if you wanted to index a weight per source field in each doc, then a 
CustomScoreQuery would certianly make more sense)


-Hoss


---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org