lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Erik Hatcher <e...@ehatchersolutions.com>
Subject Re: How to get mapping of query terms to number of their occurrences in a doc?
Date Thu, 09 Feb 2006 09:10:40 GMT
This is a real gotcha with Lucene in it's out of the box  
configuration.  In the several applications I've built to index  
documents I've always hit this and had to set the maxFieldLength to  
its maximum possible value.  Is there still an argument to be made to  
keep the default at 10K or would it be reasonable to bump this up  
even if there are, the few, cases where setting it lower is  
desirable?   We made the compound file index be the default to  
prevent file handle limitations at the expense of some (generally  
irrelevant) performance, so maybe we could also make this more common  
setting the default also?

	Erik

On Feb 8, 2006, at 2:17 PM, Dmitry Goldenberg wrote:

> Duh! Bingo! Mistery solved. I should have thought of this :)
> The discrepancies come in with larger documents, definitely > 10K  
> terms which is Lucene's default maxFieldLength.
>
> Thanks for your help, Chris
>
> - Dmitry
>
> ________________________________
>
> From: Chris Hostetter [mailto:hossman_lucene@fucit.org]
> Sent: Wed 2/8/2006 10:04 AM
> To: java-user@lucene.apache.org
> Subject: RE: How to get mapping of query terms to number of their  
> occurrences in a doc?
>
>
>
>
> : That's what I did, for debugging.  The query is "biology", and  
> here's
> : what the API tells me for term frequencies:
> : biolog 15
> : biologi 31
> : biologist 4
> :
> : I actually see 13 occurrences of "biologist" and "biologists", 64
> : occurrences of "biology", 27 occurrences of "biological".
> :
> : I see "inform 22" but the actual count of the word "information"  
> in the
> : document is 33.  But "ioniz 7" is correct.
>
> I think I missunderstood what you ment when you said the counts don't
> match up.  Are you comparing the number you get from that code with  
> the
> number of times you personally see the word in the source document  
> before
> it has been analyzed?
>
> If so, then there could be a couple of things going on ... i would  
> start
> by using a tool like Luke to see the actual lists of Terms for each  
> doc --
> there may be something else your analyzer is doing that you don't  
> realize.
>
> It's also possible that you are hitting the maxFieldLength in the
> IndexWriter ... when that happens IndexWriter throws away any  
> remaining
> tokens, so if your documenst are really large.
>
> Lastly, I would add a *lot* more debugging to your code.  Print out  
> the
> contents of "terms", when you loop over "tfvs" print out the field  
> and the
> full list of strTerms, in the inner most loop when you incriment the
> count, print out the field/text/and count.
>
> that's the best advise i have for spotting what's wrong.
>
>
>
> : ________________________________
> :
> : From: Chris Hostetter [mailto:hossman_lucene@fucit.org]
> : Sent: Tue 2/7/2006 4:10 PM
> : To: java-user@lucene.apache.org
> : Subject: Re: How to get mapping of query terms to number of their  
> occurrences in a doc?
> :
> :
> :
> :
> : A cursory reading of your code looks ok ... stemming shouldn't be  
> an issue
> : as long as your measure of success is comparing docs that match your
> : orriginal query with the counts you get out.
> :
> : What i mean by that is that any stemming should have already been  
> taken
> : care of when your query object was constructed (either by you  
> manually, or
> : by QueryParser).  the direct equals comparisons you are dong  
> should be
> : fine.
> :
> : have you tried adding logging of the raw term field/text and the  
> freq
> : counts you get back to see if that helps you spot the problem?
> :
> :
> : : Date: Mon, 6 Feb 2006 14:34:05 -0800
> : : From: Dmitry Goldenberg <dmitry.goldenberg@weblayers.com>
> : : Reply-To: java-user@lucene.apache.org
> : : To: java-user@lucene.apache.org
> : : Subject: How to get mapping of query terms to number of their  
> occurrences
> : :     in a doc?
> : :
> : : Given a query, I want to be able to, for each query term, get  
> the number of occurrences of the term.  I have tried what I'm  
> including below and it does not seem to provide reliable results.   
> Seems to work fine with exact matching but as soon as stemming  
> kicks in, all bets are off as to value of the number of occurrences  
> returned.
> : :
> : : Any ideas, anyone?  Can this be written in a simpler and/or  
> more efficient way?
> : : Thanks -
> : :
> : :       int totalOccurrences = 0;
> : :
> : :       reader = IndexReader.open(getDirectory(indexDirPath));
> : :       HashSet terms = new HashSet();
> : :       query.extractTerms(terms);
> : :
> : :       TermFreqVector[] tfvs = reader.getTermFreqVectors(docId);
> : :       if (tfvs != null) {
> : :
> : :         // For each term frequency vector (i.e. for each field)
> : :         for (int i = 0; i < tfvs.length; i++) {
> : :           String field = tfvs[i].getField();
> : :           String[] strTerms = tfvs[i].getTerms();
> : :           int[] tfs = tfvs[i].getTermFrequencies();
> : :
> : :           if (strTerms != null) {
> : :
> : :             // For each term in the query
> : :             for (Iterator iter = terms.iterator(); iter.hasNext 
> ();) {
> : :
> : :               Term term = (Term) iter.next();
> : :               // For each term in the vector
> : :               for (int j = 0; j < strTerms.length; j++) {
> : :
> : :                 // If found the query term among the vector terms
> : :                 if (field.equals(term.field()) && strTerms 
> [j].equals(term.text())) {
> : :
> : :                   // Add the term frequency to the total
> : :                   totalOccurrences += tfs[j];
> : :
> : :                 }
> : :               }
> : :             }
> : :           }
> : :         }
> : :       }
> : :
> :
> :
> :
> : -Hoss
> :
> :
> :  
> ---------------------------------------------------------------------
> : To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> : For additional commands, e-mail: java-user-help@lucene.apache.org
> :
> :
> :
> :
> :
>
>
>
> -Hoss
>
>
> ---------------------------------------------------------------------
> 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