lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Fabrício Raphael <fabricio...@gmail.com>
Subject Re: How to give a score for all documents?
Date Fri, 28 Aug 2009 18:59:14 GMT
Any ideia? Or is it a limitation of the Lucene?

2009/8/25 Fabrício Raphael <fabriciorsf@gmail.com>

> First, that a document is relevant to a query does not necessarily mean
> that this document has to contain some query term. You can have other ways
> to assert that a document is relevant to a query.
>
> My IR model is different of the vector model, so it can to give score not
> null for documents irrelevant for the vector model. I know that Lucene
> implements the vector model, but I want to use the facilities of the Lucene
> because I like what the Lucene provides.
>
> But the Lucene to give scores only for relevant documents for the vector
> model. And the my model can to give score same that this documents isn't
> relevant to vector model. It depends of the configuration granularity of
> execution.
>
> So I liked that method nextDoc() of the class that implements the
> Scorer.class returned all the documents the end of the iteration to
> calculate the score.
>
> I've got to calculate the customized score of the documents that Lucene
> returns in according to the vector model.
>
> I hope you have understood me!
>
> Thanks!
>
>
>
> On Tue, Aug 25, 2009 at 1:17 PM, Shai Erera <serera@gmail.com> wrote:
>
>> Can you please elaborate more on the use case? Why if a certain document
>> is
>> irrelevant to a certain query, you'd like to give it a score? Are you
>> perhaps talking about certain documents which should always appear in
>> search
>> results, no matter what the query is? And instead of always showing them,
>> you'd like to give them a "static score", so that they can compete w/
>> other
>> docs?
>>
>> If that's the case, I think you can use a BooleanQuery such that the user
>> query is added as a clause and then you add another clause (MUST) which is
>> in fact a MatchAllDocsQuery or something like that which returns a
>> customized score. It's expensive though as for each query you'll score all
>> docs in the index.
>>
>> But I don't think that will help (at least for this use case) since every
>> relevant document to the query will be added the same score as an
>> 'irrelevant' document, which means the relevant docs will still win, no?
>>
>> Shai
>>
>> 2009/8/25 Fabrício Raphael <fabriciorsf@gmail.com>
>>
>> > I already know about this, but I want to give a customized score for all
>> > documents in collection, independent if wache document is or isn't
>> relevant
>> > to the vector model.
>> >
>> > The similarity function is called only when the document is relevant to
>> the
>> > vector model.
>> >
>> > Do you understand me?
>> >
>> > Thanks!
>> >
>> > On Sat, Aug 22, 2009 at 2:28 AM, prashant ullegaddi <
>> > prashullegaddi@gmail.com> wrote:
>> >
>> > > If you want to modify the way Lucene scores documents, I guess you
>> need
>> > to
>> > > extend Similarity class and provide your own implementation. Take a
>> look
>> > > at:
>> > >
>> > >
>> > >
>> >
>> http://lucene.apache.org/java/2_4_1/api/org/apache/lucene/search/DefaultSimilarity.html
>> > >
>> > >
>> >
>> http://lucene.apache.org/java/2_4_1/api/org/apache/lucene/search/Similarity.html
>> > >
>> > > Thanks,
>> > > Prashant.
>> > >
>> > > 2009/8/21 Fabrício Raphael <fabriciorsf@gmail.com>
>> > >
>> > > > How to give a customize score for all documents independent of the
>> > vector
>> > > > model?
>> > > >
>> > > > I already know how to give a customize score, but I want to give
>> this
>> > > > customize score for all documents in the collection, regardless of
>> what
>> > > is
>> > > > relevant to the vector model.
>> > > >
>> > > > How to do this?
>> > > >
>> > > > Now, thanks!
>> > > >
>> > > > --
>> > > > Fabrício Raphael
>> > > >
>> > >
>> >
>> >
>> >
>> > --
>> > Fabrício Raphael
>> >
>>
>
>
>
> --
> Fabrício Raphael
>



-- 
Fabrício Raphael

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