lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Renaud Waldura" <renaud.wald...@library.ucsf.edu>
Subject RE: Wildcard query with untokenized punctuation (again)
Date Thu, 14 Jun 2007 19:50:53 GMT
Thank you for this crystal-clear explanation Mark!

> Are you sure you need a PhraseQuery and not a Boolean 
> query of Should clauses?

Excellent question. What's the requirement, hey? Well, the requirement is to
find documents referring to "annanicole smith", "anna smith" and "annaliese
smith" etc. when users enter "smith,ann*". Right now nothing is returned. 

Now, it can be argued whether it should be a phrase query or not. What does
a user expect when searching for <<smith,ann*>>? In our application, this
comma-separated format is used in documents metadata, so I think users
expect an exact match, hence a phrase query. (I'll confirm with the PM.)

> I don't think the support would be that difficult (use a wildcard 
> term enumerator to correctly fill out a MultiPhraseQuery), but it 
> might take some thought to get the QueryParser to act as you want 
> (generate a PhraseQuery or MultiPhraseQuery when it sees <<smith ann*>>).

Indeed. This is what I'm struggling with. How would I do that?

I have also found
http://www.nabble.com/Phrase-queries-with-wildcards-tf2748584.html#a7668492
which discusses precisely this issue: how to get QueryParser to generate
MultiPhraseQueries. Got some good ideas from it, but unfortunately no
complete solution. I'll keep on hacking.

--Renaud
 

-----Original Message-----
From: Mark Miller [mailto:markrmiller@gmail.com] 
Sent: Thursday, June 14, 2007 12:07 PM
To: java-user@lucene.apache.org
Subject: Re: Wildcard query with untokenized punctuation (again)

All depends on what you are looking for. Ill try and give a hint as to what
is going on now:

When the QueryParser parsers <<smith,ann>> it will shove that whole piece to
the analyzer. Your analyzer returns two tokens: smith and ann. When the
QueryParser sees that more than one token is returned from a piece that was
fed to the analyzer, it makes a PhraseQuery with the each of the returned
tokens. Remember that the QueryParser feeds the analyzer in pieces, and then
creates queries based on the number of token produced from the piece (if the
piece even goes to the analyzer).

Since you will be preprocessing the query, the query parser is going to be
parsing <<smith ann*>> which causes it to feed the analyzer smith and then
ann*...neither of these pieces produce more than one token (ann* doesnt even
go to the analyzer), so no PhraseQuery is produced. Instead you will produce
a BooleanQuery with the term smith and the wildcard query ann*, both with an
occur of whatever your default operator is.

One thing I am wondering is if you even really want the query to be a
PhraseQuery or if your just accepting the behavior you getting from the
QueryParser. Right now, PhraseQuery's do not support wildcards (nor do
MultiPhraseQuery's). I don't think the support would be that difficult (use
a wildcard term enumerator to correctly fill out a MultiPhraseQuery), but it
might take some thought to get the QueryParser to act as you want (generate
a PhraseQuery or MultiPhraseQuery when it sees <<smith ann*>>).

Are you sure you need a PhraseQuery and not a Boolean query of Should
clauses?

- Mark

On 6/14/07, Renaud Waldura <renaud.waldura@library.ucsf.edu> wrote:
>
> Thanks guys, I like it! I'm already applying some regexps before query 
> parsing anyway, so it's just another pass.
>
> Now, I'm not sure how to do that without breaking another QP feature 
> that I kind of like: the query <<smith,ann>> is parsed to 
> PhraseQuery("smith ann").
> And that seems right, from a user standpoint.
>
> In fact, considering this, I realize <<smith,ann*>> should be parsed 
> to MultiPhraseQuery("smith", "ann*"), not <<+smith +ann*>> as I said
earlier.
>
> Brrrr. Getting hairy. Any hope?
>
> --Renaud
>
>
>
> -----Original Message-----
> From: Mark Miller [mailto:markrmiller@gmail.com]
> Sent: Thursday, June 14, 2007 6:43 AM
> To: java-user@lucene.apache.org
> Subject: Re: Wildcard query with untokenized punctuation (again)
>
> Gotto agree with Erick here...best idea is just to preprocess the 
> query before sending it to the QueryParser.
>
> My first thought is always to get out the sledgehammer...
>
> - Mark
>
> Erick Erickson wrote:
> > Well, perhaps the simplest thing would be to pre-process the query 
> > and make the comma into a whitespace before sending anything to the 
> > query parser. I don't know how generalizable that sort of solution 
> > is in your problem space though....
> >
> > Best
> > Erick
> >
> > On 6/13/07, Renaud Waldura <renaud.waldura@library.ucsf.edu> wrote:
> >>
> >> My very simple analyzer produces tokens made of digits and/or 
> >> letters only.
> >> Anything else is discarded. E.g. the input "smith,anna" gets 
> >> tokenized as
> >> 2
> >> tokens, first "smith" then "anna".
> >>
> >> Say I have indexed documents that contained both "smith,anna" and 
> >> "smith,annanicole". To find them, I enter the query <<smith,ann*>>.
> >> The stock Lucene 2.0 query parser produces a PrefixQuery for the 
> >> single token "smith,ann". This token doesn't exist in my index, and 
> >> I don't get a match.
> >>
> >> I have found some references to this:
> >>
> >> http://www.nabble.com/Wildcard-query-with-untokenized-punctuation-t
> >> f3
> >> 378386
> >>
> >> .
> >> html
> >> but I don't understand how I can fix it. Comma-separated terms like 
> >> this can appear in any field; I don't think I can create an 
> >> untokenized field.
> >>
> >> Really what I would like in this case is for the comma to be 
> >> considered whitespace, and the query to be parsed to <<+smith
> >> +ann*>>. Any way I can do that?
> >>
> >> --Renaud
> >>
> >>
> >>
> >
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
>
>



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


Mime
View raw message