lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Scott Ganyo <sc...@ganyo.com>
Subject Re: Searching a URL with a PrefixQuery / Too Many Clauses (again...)
Date Thu, 28 Jul 2005 01:02:55 GMT
Chris,

How about indexing the domain as one field and each part of the path  
as separate terms in another field?  I'm sure you've probably already  
thought of doing this... and maybe discarded the idea because you'd  
lose the position information.  However, even though you can't just  
simply split the URL on '/' and shove it into the field, you can add  
the position information back into the term and then put it into the  
field.  Then, you would be able to completely ditch the prefix query  
and still retrieve the documents using the entire, ordered path in (I  
think) the most efficient way possible.

For example:

http://www2.warwick.ac.uk/fac/soc/law/ug/prospective/degrees/modules/ 
commonlaw/

becomes something like (using n/*** to identify the position):

domain: www2.warwick.ac.uk
path: 1/fac, 2/soc, 3/law, 4/ug, 5/propective, 6/degrees, 7/modules,  
8/commonlaw

And you could search based on any prefix you desired.  For example  
searching for this:

http://www2.warwick.ac.uk/fac/soc/law/*

would end up being a Lucene search that looks something like this  
(note: not query parser syntax!):

domain: www2.warwick.ac.uk AND path: 1/fac AND path: 2/soc AND path:  
3/law

Does that make sense?  Would it work for you?

S

On Jul 27, 2005, at 3:56 PM, Chris May wrote:

> Always domain + part of a path e.g.
>
> url:http://blogs.warwick.ac.uk/chrismay/*
>
> or
>
> url:http://www2.warwick.ac.uk/fac/soc/law/ug/prospective/degrees/ 
> modules/commonlaw/*
>
> or
>
> url:http://www2.warwick.ac.uk/services/its/*
>
>
> ... and so on. Part of the problem is that we may need to go an  
> arbitrary number of levels down the path to get an acceptably small  
> set of documents to start from - we couldn't impose a rule that  
> said something like 'specify the first 2 directories on the  
> path' (c.f my second example). We wouldn't need to query for the  
> same path over different domains though (e.g. url:*.warwick.ac.uk/ 
> about/* )
>
> thanks
>
> Chris
>
>
>
>
> On 27 Jul 2005, at 21:33, Erik Hatcher wrote:
>
>
>> Could you give some examples of the types of PrefixQuery's you'd  
>> like to use?   Is it always at a granularity of domain and path?   
>> Or are you wanting to do a prefix pieces of the domain and path?
>>
>>     Erik
>>
>> On Jul 27, 2005, at 3:47 PM, Chris May wrote:
>>
>>
>>
>>> First, apologies for what seems to be something of an FAQ.
>>>
>>> However, I've not been able to find an answer either in LIA or in  
>>> the relevant section of the FAQ (http://wiki.apache.org/jakarta- 
>>> lucene/LuceneFAQ#head-06fafb5d19e786a50fb3dfb8821a6af9f37aa831)
>>>
>>> My setup is as follows: I have an index of a few hundred thousand  
>>> web pages. I'd like the be able to construct queries that search  
>>> for some arbitrary text within a specified URL. Kind of like  
>>> google's syntax
>>>
>>> searchterm +site:www.foo.com/some/section
>>>
>>> So, I have the page title & content indexed, and the URL stored  
>>> as a keywords field, and I imagined that I'd be able to construct  
>>> a query something like this:
>>>
>>> String[] fields = new String[]  
>>> {DocumentFields.TITLE,DocumentFields.CONTENT};
>>> Query searchTextQuery = MultiFieldQueryParser.parse 
>>> (request.getSearchQuery(), fields, analyzer);
>>> PrefixQuery urlPrefix = new PrefixQuery(new Term 
>>> (DocumentFields.URL, request.getUrlPrefix()));
>>> hits = searcher.search(searchTextQuery, new QueryFilter(urlPrefix));
>>>
>>> However, as soon as the set of documents returned by the  
>>> prefixquery is more than a thousand or so, I get a  
>>> TooManyClausesException, as you might expect.
>>>
>>> AFAICS the solutions suggested in the FAQ don't seem to apply  
>>> here: I'm already using a Filter, and that's not helping (pace  
>>> suggestion 1), I don't think I can reduce the number of terms in  
>>> the index, else my URLs wouldn't be unique any more, and  
>>> increasing the number of clauses seems like a poor choice from a  
>>> scalability point of view - I anticipate queries that could  
>>> filter perhaps a hundred thousand documents or so.
>>>
>>> I'm guessing that it might be possible to do something smart by  
>>> splitting the URL up into multiple fields - for example, one for  
>>> the host and one for the path, or even one for the host and one  
>>> for host+path together - but I'm not clear on exactly how I'd use  
>>> the two fields, and how they'd help. Can someone enlighten me?
>>>
>>> Thanks in advance
>>>
>>> Chris
>>>
>>>
>>>
>>>
>>>
>>> -------------------------------------------------------------------- 
>>> -
>>> 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
>
>


---------------------------------------------------------------------
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