Module for comparing text

[Anonymous Monk]

Dear Monks,

I am looking for a Perl module that would let me to compare two texts and tell me how similar they are. I have searched on CPAN, but I couldn't find anything. I need to come up with some arbitrary value that would tell how similar two texts are. I was thinking of implementing it myself (splitting two versions into fixed length substrings and then comparing how many of those substrings are the same in both texts), but I am sure there should be a better algorithm and possibly a module for this. The whole purpose of this is to allow for a user to enter an answer on a web form that would allow for a certain level of difference from the answer stored in the database.

Thank you in advance

[Zaxo]

Look at Algorithm::Diff or, if the strings are short and expected to be similar, Text::Levenshtein, a more sophisticated direct measure of similarity.

A third way is to compress the two strings, and compress their concatenation. If they are similar, the concatenation will compress to not much more size than one of its parts. I don't know of a module that does that.

[tachyon]

I have been meaning to release this for a while. It does pretty much exactly what you want:

package Algorithm::HowSimilar;

use 5.006;
use strict;
use warnings;
use Algorithm::Diff qw(traverse_sequences);
use Carp;
require Exporter;
use vars qw( @ISA @EXPORT_OK $VERSION );
our @ISA = qw(Exporter);
@EXPORT_OK = qw( compare );
$VERSION = '0.01';

sub compare {
    my $is_array = ref $_[0] eq 'ARRAY' ? 1 : 0;
    my $i = 0;
    if ( $is_array ) {
        my $seq1 = $_[0];
        my $seq2 = $_[1];
        my (@match,@d1, @d2) = ((),(),());
        traverse_sequences( $seq1, $seq2, {
            MATCH     => sub { push @match, $seq1->[$_[0]] },
            DISCARD_A => sub { push @d1, $seq1->[$_[0]] },
            DISCARD_B => sub { push @d2, $seq2->[$_[1]] },
        });
        my $m1 = @match/(@match+@d1);
        my $m2 = @match/(@match+@d2);
        my $mav = ($m1+$m2)/2;
      return $mav, $m1, $m2, \@match, \@d1, \@d2;
    }
    else {
        my ( $seq1, $seq2 );
        if ( $_[2] and ref $_[2] eq 'CODE' ) {
            local $_ = $_[0]; $seq1 = &{$_[2]};
            local $_ = $_[1]; $seq2 = &{$_[2]};
            carp "Did not get an array ref from callback!\n"
                unless ref $seq1 eq 'ARRAY' and ref $seq2 eq 'ARRAY';
        }
        else {
            $seq1 = _tokenize($_[0]);
            $seq2 = _tokenize($_[1]);
        }
        my ($match,$d1, $d2) = ('','','');
        traverse_sequences( $seq1, $seq2, {
            MATCH     => sub { $match .= $seq1->[$_[0]] },
            DISCARD_A => sub { $d1 .= $seq1->[$_[0]] },
            DISCARD_B => sub { $d2 .= $seq2->[$_[1]] },
        });
        my $m1 = length($match)/(length($match)+length($d1));
        my $m2 = length($match)/(length($match)+length($d2));
        my $mav = ($m1+$m2)/2;
      return $mav, $m1, $m2, $match, $d1, $d2;
    }

}

sub _tokenize { return [split //, $_[0]] }

1;
__END__

=head1 NAME

Algorithm::HowSimilar - Perl extension for quantifying similarites bet
+ween things

=head1 SYNOPSIS

  use Algorithm::HowSimilar qw(compare);
  @res = compare( $str1, $str2, sub { s/\s+//g; [split //] } );
  @res = compare( \@ary1, \@ary2 );

=head1 DESCRIPTION

This module leverages Algorithm::Diff to let you compare the degree of
+ sameness
of array or strings. It returns a result set that defines exactly how 
+similar
these things are.

=head1 METHODS

=head2 compare( ARG1, ARG2, OPTIONAL_CALLBACK )

You can call compare with either two strings compare( $str1, $str2 ):

    my ( $av_similarity,
         $sim_str1_to_str2,
         $sim_str2_to_str1,
         $matches,
         $in_str1_but_not_str2,
         $in_str2_but_not_str1
       ) = compare( 'this is a string-a', 'this is a string bbb' );

Note that the mathematical similarities of one string to another will 
+be
different unless the strings have the same length. The first result re
+turned
is the average similarity. Totally dissimilar strings will return 0. I
+dentical
strings will return 1. The degree of similarity therefore ranges from 
+0-1 and
is reported as the biggest float your OS/Perl can manage.

You can also compare two array refs compare( \@ary1, \@ary2 ):

    my ( $av_similarity,
         $sim_ary1_to_ary2,
         $sim_ary2_to_ary1,
         $ref_ary_matches,
         $ref_ary_in_ary1_but_not_ary2,
         $ref_ary_in_ary2_but_not_ary1
       ) = compare( [ 1,2,3,4 ], [ 3,4,5,6,7 ] );


When called with two string you can specify an optional callback that 
+changes
the default tokenization of strings (a simple split on null) to whatev
+er you
need. The strings are passed to you callback in $_ and the sub is expe
+cted to
return an array ref. So for example to ignore all
whitespace you could:

    @res = compare( 'this is a string',
                    'this is a string ',
                    sub { s/\s+//g; [split //] }
                  );

You already get the intersection of the strings or arrays. You can get
+ the
union like this:

    @res = compare( $str1, $str2 );
    $intersection = $res[3];
    $union = $res[3].$res[4].$res[5];
    @res = compare( \@ary1, \@ary2 );
    @intersection = @{$res[3]};
    @union = ( @{$res[3]}, @{$res[4]}, @{$res[5]} );

=head2 EXPORT

None by default.

=head1 AUTHOR

Dr James Freeman <[e-mail]>

=head1 SEE ALSO

L<perl>.

=cut

[matthewb]

Not used it myself, but a quick search on CPAN with the keyword `diff' turns up Text::Diff , which looks like the sort of thing you are after.

MB

[ehdonhon]

Text::Diff
Algorithm::Diff

[dree]

Take also a look at Text::PhraseDistance

[castaway]

To extend the list:
String::Approx

C.