You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@jackrabbit.apache.org by Martin Perez <mp...@gmail.com> on 2006/01/21 13:37:50 UTC

Jackrabbit search score algorithm

Hi.

I'm searching some words on jackrabbit (a month ago release, sorry if this
havs changed) string properties and binary content, and every results come
with a jcr:score of 1000.

What is the followed algorithm? is that result ok? I was expecting something
like an score based on the occurrences or something similar.

Martin

Re: Jackrabbit search score algorithm

Posted by Martin Perez <mp...@gmail.com>.
Of course, JCRConstants.JCR_SCORE == jcr:score

On 1/23/06, Martin Perez <mp...@gmail.com> wrote:
>
> I'm using this query:
>
> statement = "/jcr:root"+ a path
> +"//element(*,nt:resource)[jcr:contains(.,'phrase')]"
>
> And for getting the score:
>
>             NodeIterator nodeIterator = result.getNodes();
>             while (it.hasNext()) {
>                 javax.jcr.query.Row row = (javax.jcr.query.Row) it.nextRow
> ();
>                 javax.jcr.Node node = (javax.jcr.Node
> )nodeIterator.nextNode();
>                 double score = row.getValue(JCRConstants.JCR_SCORE
> ).getDouble();
>
> I think the query is ok. In fact, returns several hits. But all of them
> have an score of 1000.
>
> Could I am getting the score wrongly? Maybe could it be due to my
> jackrabbit version?
>
> Martin
>
> On 1/23/06, Marcel Reutegger < marcel.reutegger@gmx.net> wrote:
> >
> > Jackrabbit uses the default lucene algorithm [1] to calculate the score
> > for a jcr:contains clause. any other query element will usually return a
> > score of 1000.
> >
> > a quick test showed the following for the query:
> >
> > //*[jcr:contains(.,'apache')] order by @jcr:score descending
> >
> > jcr:score  |  text property
> > ----------------------------------------------------------------------
> > 1000       | "Apache Jackrabbit"
> > 848        | "some test jackrabbit apache, apache is great"
> > 350        | "this is a text that is much larger than the first one" +
> >               "and only contains the word apache once."
> >
> > regards
> >   marcel
> >
> > [1]
> >
> > http://lucene.apache.org/java/docs/api/org/apache/lucene/search/Similarity.html
> >
> >
> > Martin Perez wrote:
> > > Hi.
> > >
> > > I'm searching some words on jackrabbit (a month ago release, sorry if
> > this
> > > havs changed) string properties and binary content, and every results
> > come
> > > with a jcr:score of 1000.
> > >
> > > What is the followed algorithm? is that result ok? I was expecting
> > something
> > > like an score based on the occurrences or something similar.
> > >
> > > Martin
> > >
> >
> >
>

Re: Jackrabbit search score algorithm

Posted by Martin Perez <mp...@gmail.com>.
Thanks for the quick fix Marcel. This afternoon I'll check if it works with
my queries. I suppose that it will work as they are simple ones.

Martin

On 1/23/06, Marcel Reutegger <ma...@gmx.net> wrote:
>
> ah, now that I'm using a more complex path constraint I also get scores
> that are always 1000.
>
> this is due to the current implementation of the descendant-or-self axis
> in jackrabbit. it does not propagate the score of the sub query.
>
> this is now fixed in svn revision: 371520
>
> thanks for reporting this issue.
>
> please note that there might be other query statements where the score
> value is not propaged to the final query result.
> e.g:
> //*[jcr:contains(.,'foo')]/bar
>
> will not return the proper score for 'foo' but just 1000 for the node
> name match 'bar'. the reason is basically efficiency. but if there is a
> need to support more sofisticated scoring for such queries we can
> implement this at the cost of slightly slower queries.
>
> regards
>   marcel
>
> Martin Perez wrote:
> > I'm using this query:
> >
> > statement = "/jcr:root"+ a path
> > +"//element(*,nt:resource)[jcr:contains(.,'phrase')]"
> >
> > And for getting the score:
> >
> >             NodeIterator nodeIterator = result.getNodes();
> >             while (it.hasNext()) {
> >                 javax.jcr.query.Row row = (javax.jcr.query.Row)
> it.nextRow
> > ();
> >                 javax.jcr.Node node = (javax.jcr.Node
> > )nodeIterator.nextNode();
> >                 double score = row.getValue(JCRConstants.JCR_SCORE
> > ).getDouble();
> >
> > I think the query is ok. In fact, returns several hits. But all of them
> have
> > an score of 1000.
> >
> > Could I am getting the score wrongly? Maybe could it be due to my
> jackrabbit
> > version?
> >
> > Martin
> >
> > On 1/23/06, Marcel Reutegger <ma...@gmx.net> wrote:
> >
> >>Jackrabbit uses the default lucene algorithm [1] to calculate the score
> >>for a jcr:contains clause. any other query element will usually return a
> >>score of 1000.
> >>
> >>a quick test showed the following for the query:
> >>
> >>//*[jcr:contains(.,'apache')] order by @jcr:score descending
> >>
> >>jcr:score  |  text property
> >>----------------------------------------------------------------------
> >>1000       | "Apache Jackrabbit"
> >>848        | "some test jackrabbit apache, apache is great"
> >>350        | "this is a text that is much larger than the first one" +
> >>              "and only contains the word apache once."
> >>
> >>regards
> >>  marcel
> >>
> >>[1]
> >>
> >>
> http://lucene.apache.org/java/docs/api/org/apache/lucene/search/Similarity.html
> >>
> >>
> >>Martin Perez wrote:
> >>
> >>>Hi.
> >>>
> >>>I'm searching some words on jackrabbit (a month ago release, sorry if
> >>
> >>this
> >>
> >>>havs changed) string properties and binary content, and every results
> >>
> >>come
> >>
> >>>with a jcr:score of 1000.
> >>>
> >>>What is the followed algorithm? is that result ok? I was expecting
> >>
> >>something
> >>
> >>>like an score based on the occurrences or something similar.
> >>>
> >>>Martin
> >>>
> >>
> >>
> >
>
>

Re: Jackrabbit search score algorithm

Posted by Marcel Reutegger <ma...@gmx.net>.
ah, now that I'm using a more complex path constraint I also get scores 
that are always 1000.

this is due to the current implementation of the descendant-or-self axis 
in jackrabbit. it does not propagate the score of the sub query.

this is now fixed in svn revision: 371520

thanks for reporting this issue.

please note that there might be other query statements where the score 
value is not propaged to the final query result.
e.g:
//*[jcr:contains(.,'foo')]/bar

will not return the proper score for 'foo' but just 1000 for the node 
name match 'bar'. the reason is basically efficiency. but if there is a 
need to support more sofisticated scoring for such queries we can 
implement this at the cost of slightly slower queries.

regards
  marcel

Martin Perez wrote:
> I'm using this query:
> 
> statement = "/jcr:root"+ a path
> +"//element(*,nt:resource)[jcr:contains(.,'phrase')]"
> 
> And for getting the score:
> 
>             NodeIterator nodeIterator = result.getNodes();
>             while (it.hasNext()) {
>                 javax.jcr.query.Row row = (javax.jcr.query.Row) it.nextRow
> ();
>                 javax.jcr.Node node = (javax.jcr.Node
> )nodeIterator.nextNode();
>                 double score = row.getValue(JCRConstants.JCR_SCORE
> ).getDouble();
> 
> I think the query is ok. In fact, returns several hits. But all of them have
> an score of 1000.
> 
> Could I am getting the score wrongly? Maybe could it be due to my jackrabbit
> version?
> 
> Martin
> 
> On 1/23/06, Marcel Reutegger <ma...@gmx.net> wrote:
> 
>>Jackrabbit uses the default lucene algorithm [1] to calculate the score
>>for a jcr:contains clause. any other query element will usually return a
>>score of 1000.
>>
>>a quick test showed the following for the query:
>>
>>//*[jcr:contains(.,'apache')] order by @jcr:score descending
>>
>>jcr:score  |  text property
>>----------------------------------------------------------------------
>>1000       | "Apache Jackrabbit"
>>848        | "some test jackrabbit apache, apache is great"
>>350        | "this is a text that is much larger than the first one" +
>>              "and only contains the word apache once."
>>
>>regards
>>  marcel
>>
>>[1]
>>
>>http://lucene.apache.org/java/docs/api/org/apache/lucene/search/Similarity.html
>>
>>
>>Martin Perez wrote:
>>
>>>Hi.
>>>
>>>I'm searching some words on jackrabbit (a month ago release, sorry if
>>
>>this
>>
>>>havs changed) string properties and binary content, and every results
>>
>>come
>>
>>>with a jcr:score of 1000.
>>>
>>>What is the followed algorithm? is that result ok? I was expecting
>>
>>something
>>
>>>like an score based on the occurrences or something similar.
>>>
>>>Martin
>>>
>>
>>
> 


Re: Jackrabbit search score algorithm

Posted by Martin Perez <mp...@gmail.com>.
I'm using this query:

statement = "/jcr:root"+ a path
+"//element(*,nt:resource)[jcr:contains(.,'phrase')]"

And for getting the score:

            NodeIterator nodeIterator = result.getNodes();
            while (it.hasNext()) {
                javax.jcr.query.Row row = (javax.jcr.query.Row) it.nextRow
();
                javax.jcr.Node node = (javax.jcr.Node
)nodeIterator.nextNode();
                double score = row.getValue(JCRConstants.JCR_SCORE
).getDouble();

I think the query is ok. In fact, returns several hits. But all of them have
an score of 1000.

Could I am getting the score wrongly? Maybe could it be due to my jackrabbit
version?

Martin

On 1/23/06, Marcel Reutegger <ma...@gmx.net> wrote:
>
> Jackrabbit uses the default lucene algorithm [1] to calculate the score
> for a jcr:contains clause. any other query element will usually return a
> score of 1000.
>
> a quick test showed the following for the query:
>
> //*[jcr:contains(.,'apache')] order by @jcr:score descending
>
> jcr:score  |  text property
> ----------------------------------------------------------------------
> 1000       | "Apache Jackrabbit"
> 848        | "some test jackrabbit apache, apache is great"
> 350        | "this is a text that is much larger than the first one" +
>               "and only contains the word apache once."
>
> regards
>   marcel
>
> [1]
>
> http://lucene.apache.org/java/docs/api/org/apache/lucene/search/Similarity.html
>
>
> Martin Perez wrote:
> > Hi.
> >
> > I'm searching some words on jackrabbit (a month ago release, sorry if
> this
> > havs changed) string properties and binary content, and every results
> come
> > with a jcr:score of 1000.
> >
> > What is the followed algorithm? is that result ok? I was expecting
> something
> > like an score based on the occurrences or something similar.
> >
> > Martin
> >
>
>

Re: Jackrabbit search score algorithm

Posted by Marcel Reutegger <ma...@gmx.net>.
Jackrabbit uses the default lucene algorithm [1] to calculate the score 
for a jcr:contains clause. any other query element will usually return a 
score of 1000.

a quick test showed the following for the query:

//*[jcr:contains(.,'apache')] order by @jcr:score descending

jcr:score  |  text property
----------------------------------------------------------------------
1000       | "Apache Jackrabbit"
848        | "some test jackrabbit apache, apache is great"
350        | "this is a text that is much larger than the first one" +
              "and only contains the word apache once."

regards
  marcel

[1] 
http://lucene.apache.org/java/docs/api/org/apache/lucene/search/Similarity.html


Martin Perez wrote:
> Hi.
> 
> I'm searching some words on jackrabbit (a month ago release, sorry if this
> havs changed) string properties and binary content, and every results come
> with a jcr:score of 1000.
> 
> What is the followed algorithm? is that result ok? I was expecting something
> like an score based on the occurrences or something similar.
> 
> Martin
>