lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Erick Erickson" <erickerick...@gmail.com>
Subject Re: Phrase search using quotes -- special Tokenizer
Date Sun, 03 Sep 2006 14:45:33 GMT
Disclaimer: Of course I'm not as familiar with your problem space as you
are, so I may be way out in left field, but...

I *still* think you're making waaaaaay too much work for yourself and need
to examine your assumptions.

0> But when you index something UN_TOKENIZED as in your example, I don't
think you'll  find the words "phrase" and "test" if you just search for them
individually.

1> "doesn't parse apart phrases". Why do you care? The "usual" way of
handling this is to just go ahead and parse them apart, then submit your
query with quotes embedded. Would that serve your needs? If not, I bet you
could make a cool regex that would do this for you and use a
PatternAnalyzer. If neither of those work, instead of hacking Lucene code,
make your own tokenizer by overriding one of the Lucene tokenizers. See
Lucene in Action for example, the section on synonyms as I remember.....
It'll be waaaaay less work in the long run than having to try to stay in
synch by hacking the Lucene code. Especially for the next poor soul who has
to maintain it........

2> doesn't parse/separate underscores and hyphens. Why not use a
PatternAnalyzer? You can make it do most anything you want with a clever
regex.

3> the other thing that might be producing unexpected results is that the
default is OR for QueryParser.......

Since no amount of documentation replaces a program, I've included one
illustrating what I'm talking about. It's less likely we're talk past each
other this way. All it may *really* demonstrate is that I don't understand
what you're trying to do at all, but at least we'll know <G>...

Notice in particular that it finds the hypenated words, but doesn't find
their individual parts. Also,  quoted phrasew are found, even though the
stop words aren't in the index (examined via luke). I used Eclipse to
compile/run it......

Best
Erick

package scratch;

import java.io.IOException;
import java.util.regex.Pattern;

import org.apache.lucene.analysis.Analyzer;
import org.apache.lucene.analysis.StopFilter;
import org.apache.lucene.index.memory.PatternAnalyzer;
import org.apache.lucene.document.Document;
import org.apache.lucene.document.Field;
import org.apache.lucene.index.IndexWriter;
import org.apache.lucene.queryParser.QueryParser;
import org.apache.lucene.search.Hits;
import org.apache.lucene.search.IndexSearcher;
import org.apache.lucene.search.Query;

public class Scratch {

    private Analyzer analyzer = null;

    private Analyzer getAnalyzer() {
        if (analyzer == null) {
            // Break on whitespace or anything that is NOT hyphen,
underscore
            // or letter.
            Pattern pat = Pattern.compile("(\\s|[^-_a-zA-Z])");
            analyzer = new PatternAnalyzer(pat, true,
                    StopFilter.makeStopSet(STOP_WORDS));
        }
        return analyzer;

    }

    private void makeTestIndex() throws Exception {
        IndexWriter writer = new IndexWriter("C:/mydir", getAnalyzer(),
true);
        String text = ("this is the test text hyphenated-Iamhyphenated
underscored_IaMunDerScoRed");
        Document doc = new Document();
        doc = new Document();
        doc.add(new Field("test", text, Field.Store.YES,
                        Field.Index.TOKENIZED));
        writer.addDocument(doc);

        writer.close();
    }

    private void doSearch(String query, int expectedHits) throws Exception {
        try {
            QueryParser qp = new QueryParser("test", getAnalyzer());
            qp.enable_tracing();
            IndexSearcher srch = new IndexSearcher("C:/mydir");
            Query tmp = qp.parse(query);
            // Uncomment to see parsed form of query
            // System.out.println("Parsed form is '" + tmp.toString() +
"'");
            Hits hits = srch.search(tmp);

            String msg = "";

            if (hits.length() == expectedHits) {
                msg = "Test passed ";
            } else {
                msg = "************TEST FAILED************ ";
            }
            System.out.println(msg + "Expected "
                    + Integer.toString(expectedHits) + " hits, got "
                    + Integer.toString(hits.length()) + " hits");

        } catch (IOException e) {
            System.out.println("Caught IOException");
            e.printStackTrace();
        }
    }

    private void doSearchPhrase(String phrase, int expectedHits) throws
Exception {
        doSearch("\"" + phrase + "\"", expectedHits);
    }
    public static void main(String[] args) {
        try {
            Scratch scratch = new Scratch();
            scratch.getAnalyzer();
            scratch.makeTestIndex();
            scratch.doSearch("underscored_iamunderscored", 1);
            scratch.doSearch("underscored iamunderscored", 0);
            scratch.doSearch("hyphenated-iamhyphenated", 1);
            scratch.doSearch("iamunderscored", 0);
            scratch.doSearch("underscored", 0);

            scratch.doSearchPhrase("this is the test text", 1);
            scratch.doSearchPhrase("text with hyphenated-iamhyphenated", 1);
            scratch.doSearchPhrase("text with underscored_iamunderscored",
0);


        } catch (Exception e) {
            System.err.println(e.getMessage());
        }
    }

    protected static final String[] STOP_WORDS = { "a", "an", "and", "are",
            "as", "at", "b", "be", "but", "by", "c", "d", "e", "f", "for",
            "if", "g", "h", "i", "in", "into", "is", "it", "j", "k", "l",
"m",
            "n", "no", "not", "o", "of", "on", "or", "p", "q", "r", "s",
            "such", "t", "that", "the", "their", "then", "there", "these",
            "they", "this", "to", "u", "v", "w", "was", "will", "with", "x",
            "y", "z" };

}



On 9/2/06, Philip Brown <pmb@us.ibm.com> wrote:
>
>
> I tend to agree with Mark.  I tried a query as so...
>
>    TermQuery query = new TermQuery(new Term("keywordField", "phrase
> test"));
>    IndexSearcher searcher= new IndexSearcher(activeIdx);
>    Hits hits = searcher.search(query);
>
> And this produced the expected results.  When building the index, I did
> NOT
> enclose the keywords in quotes -- just added as UN_TOKENIZED.
>
> Philip
>
>
> Mark Miller-5 wrote:
> >
> > I think if he wants to use the queryparser to parse his search strings
> > that he has no choice but to modify it. It will eat any pair of quotes
> > going through it no matter what analyzer is used.
> >
> > - Mark
> >> Well, you're flying blind. Is the behavior rooted in the indexing or
> >> querying? Since you can't answer that, you're reduced to trying random
> >> things hoping that one of them works. A little like voodoo. I've wasted
> >> faaaaarrrrrr too much time trying to solve what I was *sure* was the
> >> problem
> >> only to find it was somewhere else (the last place I look, of course)
> >> <G>...
> >>
> >> Using Luke on a RAMDir. No, I don't know how to, but it should be a
> >> simple
> >> thing to write the index to an FSDir at the same time you create your
> >> RAMDir
> >> and use Luke then. This is debugging, after all.
> >>
> >> I'd be really, really, really reluctant to modify the query parser
> and/or
> >> the tokenizer, since whenever I've been tempted it's usually because I
> >> don't
> >> understand the tools already provided. Then I have to maintain my
> custom
> >> code. Which sucks. Although it sure feels more productive to hack a
> >> bunch of
> >> code and get something that works 90% of the time, then spend weeks
> >> making
> >> the other 10% work than taking two days to find the 3 lines you
> *really*
> >> need <G>.
> >>
> >> Have you thought of a PatternAnalyzer? It takes a regular expression
> >> as the
> >> tokenizer and  (from the Javadoc)
> >> <<< Efficient Lucene analyzer/tokenizer that preferably operates on
a
> >> String
> >> rather than a
> >> Reader<http://java.sun.com/j2se/1.4/docs/api/java/io/Reader.html>,
> >> that can flexibly separate text into terms via a regular expression
> >> Pattern<
> http://java.sun.com/j2se/1.4/docs/api/java/util/regex/Pattern.html>(with
> >>
> >> behaviour identical to
> >> String.split(String)<
> http://java.sun.com/j2se/1.4/docs/api/java/lang/String.html#split%28java.lang.String%29
> >),
> >>
> >> and that combines the functionality of
> >> LetterTokenizer<
> file:///C:/lucene_1.9.1/docs/api/org/apache/lucene/analysis/LetterTokenizer.html
> >,
> >>
> >> LowerCaseTokenizer<
> file:///C:/lucene_1.9.1/docs/api/org/apache/lucene/analysis/LowerCaseTokenizer.html
> >,
> >>
> >> WhitespaceTokenizer<
> file:///C:/lucene_1.9.1/docs/api/org/apache/lucene/analysis/WhitespaceTokenizer.html
> >,
> >>
> >> StopFilter<
> file:///C:/lucene_1.9.1/docs/api/org/apache/lucene/analysis/StopFilter.html
> >into
> >>
> >> a single efficient multi-purpose class.>>>
> >>
> >> One word of caution, the regular expression consists of expressions
> that
> >> *break* tokens, not expressions that *form* words, which threw me at
> >> first.
> >> Just like the doc says, like splitstring <G>.... This is in 2.0,
> >> although I
> >> *believe* it's also in the contrib section of 1.9 (or is in the
> >> regular API,
> >> I forget).
> >>
> >> Best
> >> Erick
> >>
> >> On 9/1/06, Philip Brown <pmb@us.ibm.com> wrote:
> >>>
> >>>
> >>> No, I've never used Luke.  Is there an easy way to examine my
> >>> RAMDirectory
> >>> index?  I can create the index with no quoted keywords, and when I
> >>> search
> >>> for a keyword, I get back the expected results (just can't search for
> a
> >>> phrase that has whitespace in it).  If I create the index with
> >>> phrases in
> >>> quotes, then when I search for anything in double quotes, I get back
> >>> nothing.  If I create the index with everything in quotes, then when I
> >>> search for anything by the keyword field, I get nothing, regardless of
> >>> whether I use quotes in the query string or not.  (I can get results
> >>> back
> >>> by
> >>> searching on other fields.)  What do you think?
> >>>
> >>> Philip
> >>>
> >>>
> >>> Erick Erickson wrote:
> >>> >
> >>> > OK, I've gotta ask. Have you examined your index with Luke to see if
> >>> what
> >>> > you *think* is in the index actually *is*???
> >>> >
> >>> > Erick
> >>> >
> >>> > On 9/1/06, Philip Brown <pmb@us.ibm.com> wrote:
> >>> >>
> >>> >>
> >>> >> Interesting...just ran a test where I put double quotes around
> >>> everything
> >>> >> (including single keywords) of source text and then ran searches
> >>> for a
> >>> >> known
> >>> >> keyword with and without double quotes -- doesn't find either time.
> >>> >>
> >>> >>
> >>> >> Mark Miller-5 wrote:
> >>> >> >
> >>> >> > Sorry to hear you're having trouble. You indeed need the double
> >>> quotes
> >>> >> in
> >>> >> > the source text. You will also need them in the query string.
> Make
> >>> sure
> >>> >> > they
> >>> >> > are in both places. My machine is hosed right now or I would
do
> it
> >>> for
> >>> >> you
> >>> >> > real quick. My guess is that I forgot to mention...no only
do you
> >>> need
> >>> >> to
> >>> >> > add the <QUOTED> definiton to the TOKEN section, but
below that
> you
> >>> >> will
> >>> >> > find the grammer...you need to add <QUOTED> to the grammer.
If
> you
> >>> look
> >>> >> > how
> >>> >> > <NUM> and <APOSTROPHE> are done you will prob
see what you
> >>> should do.
> >>> >> If
> >>> >> > not, my machine should be back up tomarrow...
> >>> >> >
> >>> >> > - Mark
> >>> >> >
> >>> >> > On 9/1/06, Philip Brown <pmb@us.ibm.com> wrote:
> >>> >> >>
> >>> >> >>
> >>> >> >> Well, I tried that, and it doesn't seem to work still.
 I would
> be
> >>> >> happy
> >>> >> >> to
> >>> >> >> zip up the new files, so you can see what I'm using --
maybe
> >>> you can
> >>> >> get
> >>> >> >> it
> >>> >> >> to work.  The first time, I tried building the documents
without
> >>> >> quotes
> >>> >> >> surrounding each phrase.  Then, I retried by enclosing
every
> >>> phrase
> >>> >> >> within
> >>> >> >> double quotes.  Neither seemed to work.  When constructing
the
> >>> query
> >>> >> >> string
> >>> >> >> for the search, I always added the double quotes (otherwise,
> it'd
> >>> >> think
> >>> >> >> it
> >>> >> >> was multiple terms).  (I didn't even test the underscore
and
> >>> >> hyphenated
> >>> >> >> terms.)  I thought Lucene was (sort of by default) set
up to
> >>> search
> >>> >> >> quoted
> >>> >> >> phrases.  From
> >>> http://lucene.apache.org/java/docs/api/index.html -->
> >>> A
> >>> >> >> Phrase is a group of words surrounded by double quotes
such as
> >>> "hello
> >>> >> >> dolly".  So, this should be easy, right?  I must be missing
> >>> something
> >>> >> >> stupid.
> >>> >> >>
> >>> >> >> Thanks,
> >>> >> >>
> >>> >> >> Philip
> >>> >> >>
> >>> >> >>
> >>> >> >> Mark Miller-5 wrote:
> >>> >> >> >
> >>> >> >> > So this will recognize anything in quotes as a single
token
> and
> >>> '_'
> >>> >> and
> >>> >> >> > '-' will not break up words. There may be some repercussions
> for
> >>> the
> >>> >> >> NUM
> >>> >> >> > token but nothing I'd worry about. maybe you want
to use
> Unicode
> >>> for
> >>> >> >> '-'
> >>> >> >> > and '_' as well...I wouldn't worry about it myself.
> >>> >> >> >
> >>> >> >> > - Mark
> >>> >> >> >
> >>> >> >> >
> >>> >> >> > TOKEN : {                      // token patterns
> >>> >> >> >
> >>> >> >> >   // basic word: a sequence of digits & letters
> >>> >> >> >   <ALPHANUM: (<LETTER>|<DIGIT>|<KOREAN>)+
>
> >>> >> >> >
> >>> >> >> > | <QUOTED:     "\"" (~["\""])+ "\"">
> >>> >> >> >
> >>> >> >> >   // internal apostrophes: O'Reilly, you're, O'Reilly's
> >>> >> >> >   // use a post-filter to remove possesives
> >>> >> >> > | <APOSTROPHE: <ALPHA> ("'" <ALPHA>)+
>
> >>> >> >> >
> >>> >> >> >   // acronyms: U.S.A., I.B.M., etc.
> >>> >> >> >   // use a post-filter to remove dots
> >>> >> >> > | <ACRONYM: <ALPHA> "." (<ALPHA> ".")+
>
> >>> >> >> >
> >>> >> >> >   // company names like AT&T and Excite@Home.
> >>> >> >> > | <COMPANY: <ALPHA> ("&"|"@") <ALPHA>
>
> >>> >> >> >
> >>> >> >> >   // email addresses
> >>> >> >> > | <EMAIL: <ALPHANUM> (("."|"-"|"_") <ALPHANUM>)*
"@"
> <ALPHANUM>
> >>> >> >> > (("."|"-") <ALPHANUM>)+ >
> >>> >> >> >
> >>> >> >> >   // hostname
> >>> >> >> > | <HOST: <ALPHANUM> ("." <ALPHANUM>)+
>
> >>> >> >> >
> >>> >> >> >   // floating point, serial, model numbers, ip addresses,
etc.
> >>> >> >> >   // every other segment must have at least one digit
> >>> >> >> > | <NUM: (<ALPHANUM> <P> <HAS_DIGIT>
> >>> >> >> >        | <HAS_DIGIT> <P> <ALPHANUM>
> >>> >> >> >        | <ALPHANUM> (<P> <HAS_DIGIT>
<P> <ALPHANUM>)+
> >>> >> >> >        | <HAS_DIGIT> (<P> <ALPHANUM>
<P> <HAS_DIGIT>)+
> >>> >> >> >        | <ALPHANUM> <P> <HAS_DIGIT>
(<P> <ALPHANUM> <P>
> >>> >> <HAS_DIGIT>)+
> >>> >> >> >        | <HAS_DIGIT> <P> <ALPHANUM>
(<P> <HAS_DIGIT> <P>
> >>> >> <ALPHANUM>)+
> >>> >> >> >         )
> >>> >> >> >   >
> >>> >> >> > | <#P: ("_"|"-"|"/"|"."|",") >
> >>> >> >> > | <#HAS_DIGIT:                      // at least
one digit
> >>> >> >> >     (<LETTER>|<DIGIT>)*
> >>> >> >> >     <DIGIT>
> >>> >> >> >     (<LETTER>|<DIGIT>)*
> >>> >> >> >   >
> >>> >> >> >
> >>> >> >> > | < #ALPHA: (<LETTER>)+>
> >>> >> >> > | < #LETTER:                      // unicode letters
> >>> >> >> >       [
> >>> >> >> >        "\u0041"-"\u005a",
> >>> >> >> >        "\u0061"-"\u007a",
> >>> >> >> >        "\u00c0"-"\u00d6",
> >>> >> >> >        "\u00d8"-"\u00f6",
> >>> >> >> >        "\u00f8"-"\u00ff",
> >>> >> >> >        "\u0100"-"\u1fff",
> >>> >> >> >        "-", "_"
> >>> >> >> >       ]
> >>> >> >> >   >
> >>> >> >> >
> >>> >> >> >
> >>> >>
> ---------------------------------------------------------------------
> >>> >> >> > To unsubscribe, e-mail:
> java-user-unsubscribe@lucene.apache.org
> >>> >> >> > For additional commands, e-mail:
> >>> java-user-help@lucene.apache.org
> >>> >> >> >
> >>> >> >> >
> >>> >> >> >
> >>> >> >>
> >>> >> >> --
> >>> >> >> View this message in context:
> >>> >> >>
> >>> >>
> >>>
> http://www.nabble.com/Phrase-search-using-quotes----special-Tokenizer-tf2200760.html#a6106920
> >>>
> >>> >> >> Sent from the Lucene - Java Users forum at Nabble.com.
> >>> >> >>
> >>> >> >>
> >>> >> >>
> >>> ---------------------------------------------------------------------
> >>> >> >> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> >>> >> >> For additional commands, e-mail:
> java-user-help@lucene.apache.org
> >>> >> >>
> >>> >> >>
> >>> >> >
> >>> >> >
> >>> >>
> >>> >> --
> >>> >> View this message in context:
> >>> >>
> >>>
> http://www.nabble.com/Phrase-search-using-quotes----special-Tokenizer-tf2200760.html#a6107649
> >>>
> >>> >> Sent from the Lucene - Java Users forum at Nabble.com.
> >>> >>
> >>> >>
> >>> >>
> ---------------------------------------------------------------------
> >>> >> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> >>> >> For additional commands, e-mail: java-user-help@lucene.apache.org
> >>> >>
> >>> >>
> >>> >
> >>> >
> >>>
> >>> --
> >>> View this message in context:
> >>>
> http://www.nabble.com/Phrase-search-using-quotes----special-Tokenizer-tf2200760.html#a6109067
> >>>
> >>> Sent from the Lucene - Java Users forum at Nabble.com.
> >>>
> >>>
> >>> ---------------------------------------------------------------------
> >>> 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
> >
> >
> >
>
> --
> View this message in context:
> http://www.nabble.com/Phrase-search-using-quotes----special-Tokenizer-tf2200760.html#a6115360
> Sent from the Lucene - Java Users forum at Nabble.com.
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
>
>

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