package KinoSearch::Highlight::Highlighter;
use strict;
use warnings;
use KinoSearch::Util::ToolSet;
use base qw( KinoSearch::Util::Class );
use locale;
BEGIN {
__PACKAGE__->init_instance_vars(
# constructor params / members
excerpt_field => undef,
analyzer => undef,
formatter => undef,
encoder => undef,
terms => undef,
excerpt_length => 200,
pre_tag => undef, # back compat
post_tag => undef, # back compat
token_re => qr/\b\w+(?:'\w+)?\b/,
# members
limit => undef,
);
__PACKAGE__->ready_get_set(qw( terms ));
}
use KinoSearch::Highlight::SimpleHTMLFormatter;
use KinoSearch::Highlight::SimpleHTMLEncoder;
sub init_instance {
my $self = shift;
croak("Missing required arg 'excerpt_field'")
unless defined $self->{excerpt_field};
$self->{terms} = [];
# assume HTML
if ( !defined $self->{encoder} ) {
$self->{encoder} = KinoSearch::Highlight::SimpleHTMLEncoder->new;
}
if ( !defined $self->{formatter} ) {
my ( $pre_tag, $post_tag ) = @{$self}{qw( pre_tag post_tag )};
$pre_tag = '<strong>' unless defined $pre_tag;
$post_tag = '</strong>' unless defined $post_tag;
$self->{formatter} = KinoSearch::Highlight::SimpleHTMLFormatter->new(
pre_tag => $pre_tag,
post_tag => $post_tag,
);
}
# scoring window is 1.66 * excerpt_length, with the loc in the middle
$self->{limit} = int( $self->{excerpt_length} / 3 );
}
sub generate_excerpt {
my ( $self, $doc ) = @_;
my $excerpt_length = $self->{excerpt_length};
my $limit = $self->{limit};
my $token_re = $self->{token_re};
# retrieve the text from the chosen field
my $field = $doc->get_field( $self->{excerpt_field} );
my $text = $field->get_value;
my $text_length = bytes::length $text;
return '' unless $text_length;
# determine the rough boundaries of the excerpt
my $posits = $self->_starts_and_ends($field);
my $best_location = $self->_calc_best_location($posits);
my $top = $best_location - $limit;
# expand the excerpt if the best location is near the end
$top =
$text_length - $excerpt_length < $top
? $text_length - $excerpt_length
: $top;
# if the best starting point is the very beginning, cool...
if ( $top <= 0 ) {
$top = 0;
}
# ... otherwise ...
else {
# lop off $top bytes
$text = bytes::substr( $text, $top );
# try to start the excerpt at a sentence boundary
if ($text =~ s/
\A
(
\C{0,$limit}?
\.\s+
)
//xsm
)
{
$top += bytes::length($1);
}
# no sentence boundary, so we'll need an ellipsis
else {
# skip past possible partial tokens, prepend an ellipsis
if ($text =~ s/
\A
(
\C{0,$limit}? # don't go outside the window
$token_re # match possible partial token
.*? # ... and any junk following that token
)
(?=$token_re) # just before the start of a full token...
/... /xsm # ... insert an ellipsis
)
{
$top += bytes::length($1);
$top -= 4 # three dots and a space
}
}
}
# remove possible partial tokens from the end of the excerpt
$text = bytes::substr( $text, 0, $excerpt_length + 1 );
if ( bytes::length($text) > $excerpt_length ) {
my $extra_char = chop $text;
# if the extra char wasn't part of a token, we aren't splitting one
if ( $extra_char =~ $token_re ) {
$text =~ s/$token_re$//; # if this is unsuccessful, that's fine
}
}
# if the excerpt doesn't end with a full stop, end with an an ellipsis
if ( $text !~ /\.\s*\Z/xsm ) {
$text =~ s/\W+\Z//xsm;
while ( bytes::length($text) + 4 > $excerpt_length ) {
my $extra_char = chop $text;
if ( $extra_char =~ $token_re ) {
$text =~ s/\W+$token_re\Z//xsm; # if unsuccessful, that's fine
}
$text =~ s/\W+\Z//xsm;
}
$text .= ' ...';
}
# remap locations now that we know the starting and ending bytes
$text_length = bytes::length($text);
my @relative_starts = map { $_->[0] - $top } @$posits;
my @relative_ends = map { $_->[1] - $top } @$posits;
# get rid of pairs with at least one member outside the text
while ( @relative_starts and $relative_starts[0] < 0 ) {
shift @relative_starts;
shift @relative_ends;
}
while ( @relative_ends and $relative_ends[-1] > $text_length ) {
pop @relative_starts;
pop @relative_ends;
}
# insert highlight tags
my $formatter = $self->{formatter};
my $encoder = $self->{encoder};
my $output_text = '';
my ( $start, $end, $last_start, $last_end ) = ( undef, undef, 0, 0 );
while (@relative_starts) {
$end = shift @relative_ends;
$start = shift @relative_starts;
$output_text .= $encoder->encode(
bytes::substr( $text, $last_end, $start - $last_end ) );
$output_text .= $formatter->highlight(
$encoder->encode( bytes::substr( $text, $start, $end - $start ) )
);
$last_end = $end;
}
$output_text .= $encoder->encode( bytes::substr( $text, $last_end ) );
return $output_text;
}
=for comment
Find all points in the text where a relevant term begins and ends. For terms
that are part of a phrase, only include points that are part of the phrase.
=cut
sub _starts_and_ends {
my ( $self, $field ) = @_;
my @posits;
my %done;
TERM: for my $term ( @{ $self->{terms} } ) {
if ( a_isa_b( $term, 'KinoSearch::Index::Term' ) ) {
my $term_text = $term->get_text;
next TERM if $done{$term_text};
$done{$term_text} = 1;
# add all starts and ends
my $term_vector = $field->term_vector($term_text);
next TERM unless defined $term_vector;
my $starts = $term_vector->get_start_offsets;
my $ends = $term_vector->get_end_offsets;
while (@$starts) {
push @posits, [ shift @$starts, shift @$ends, 1 ];
}
}
# intersect positions for phrase terms
else {
# if not a Term, it's an array of Terms representing a phrase
my @term_texts = map { $_->get_text } @$term;
my $phrase_text = join( ' ', @term_texts );
next TERM if $done{$phrase_text};
$done{$phrase_text} = 1;
my $posit_vec = KinoSearch::Util::BitVector->new;
my @term_vectors = map { $field->term_vector($_) } @term_texts;
# make sure all terms are present
next TERM unless scalar @term_vectors == scalar @term_texts;
my $i = 0;
for my $tv (@term_vectors) {
# one term missing, ergo no phrase
next TERM unless defined $tv;
if ( $i == 0 ) {
$posit_vec->set( @{ $tv->get_positions } );
}
else {
# filter positions using logical "and"
my $other_posit_vec = KinoSearch::Util::BitVector->new;
$other_posit_vec->set(
grep { $_ >= 0 }
map { $_ - $i } @{ $tv->get_positions }
);
$posit_vec->logical_and($other_posit_vec);
}
$i++;
}
# add only those starts/ends that belong to a valid position
my $tv_start_positions = $term_vectors[0]->get_positions;
my $tv_starts = $term_vectors[0]->get_start_offsets;
my $tv_end_positions = $term_vectors[-1]->get_positions;
my $tv_ends = $term_vectors[-1]->get_end_offsets;
$i = 0;
my $j = 0;
my $last_token_index = $#term_vectors;
for my $valid_position ( @{ $posit_vec->to_arrayref } ) {
while ( $i <= $#$tv_start_positions ) {
last if ( $tv_start_positions->[$i] >= $valid_position );
$i++;
}
$valid_position += $last_token_index;
while ( $j <= $#$tv_end_positions ) {
last if ( $tv_end_positions->[$j] >= $valid_position );
$j++;
}
push @posits,
[ $tv_starts->[$i], $tv_ends->[$j], scalar @$term ];
$i++;
$j++;
}
}
}
# sort, uniquify and return
@posits = sort { $a->[0] <=> $b->[0] || $b->[1] <=> $a->[1] } @posits;
my @unique;
my $last = ~0;
for (@posits) {
push @unique, $_ if $_->[0] != $last;
$last = $_->[0];
}
return \@unique;
}
=for comment
Select the byte address representing the greatest keyword density. Because
the algorithm counts bytes rather than characters, it will degrade if the
number of bytes per character is larger than 1.
=cut
sub _calc_best_location {
my ( $self, $posits ) = @_;
my $window = $self->{limit} * 2;
# if there aren't any keywords, take the excerpt from the top of the text
return 0 unless @$posits;
my %locations = map { ( $_->[0] => 0 ) } @$posits;
# if another keyword is in close proximity, add to the loc's score
for my $loc_index ( 0 .. $#$posits ) {
# only score positions that are in range
my $location = $posits->[$loc_index][0];
my $other_loc_index = $loc_index - 1;
while ( $other_loc_index > 0 ) {
my $diff = $location - $posits->[$other_loc_index][0];
last if $diff > $window;
my $num_tokens_at_pos = $posits->[$other_loc_index][2];
$locations{$location}
+= ( 1 / ( 1 + log($diff) ) ) * $num_tokens_at_pos;
--$other_loc_index;
}
$other_loc_index = $loc_index + 1;
while ( $other_loc_index <= $#$posits ) {
my $diff = $posits->[$other_loc_index] - $location;
last if $diff > $window;
my $num_tokens_at_pos = $posits->[$other_loc_index][2];
$locations{$location}
+= ( 1 / ( 1 + log($diff) ) ) * $num_tokens_at_pos;
++$other_loc_index;
}
}
# return the highest scoring position
return ( sort { $locations{$b} <=> $locations{$a} } keys %locations )[0];
}
1;
__END__
=head1 NAME
KinoSearch::Highlight::Highlighter - create and highlight excerpts
=head1 SYNOPSIS
my $highlighter = KinoSearch::Highlight::Highlighter->new(
excerpt_field => 'bodytext',
);
$hits->create_excerpts( highlighter => $highlighter );
=head1 DESCRIPTION
KinoSearch's Highlighter can be used to select a relevant snippet from a
document, and to surround search terms with highlighting tags. It handles
both stems and phrases correctly and efficiently, using special-purpose data
generated at index-time.
=head1 METHODS
=head2 new
my $highlighter = KinoSearch::Highlight::Highlighter->new(
excerpt_field => 'bodytext', # required
excerpt_length => 150, # default: 200
formatter => $formatter, # default: SimpleHTMLFormatter
encoder => $encoder, # default: SimpleHTMLEncoder
);
Constructor. Takes hash-style parameters:
=over
=item *
B<excerpt_field> - the name of the field from which to draw the excerpt. This
field B<must> be C<vectorized>.
=item *
B<excerpt_length> - the length of the excerpt, in I<bytes>. This should
probably use characters as a unit instead of bytes, and the behavior is likely
to change in the future.
=item *
B<formatter> - an object which subclasses L<KinoSearch::Highlight::Formatter>,
used to perform the actual highlighting.
=item *
B<encoder> - an object which subclasses L<KinoSearch::Highlight::Encoder>.
All excerpt text gets passed through the encoder, including highlighted terms.
By default, this is a SimpleHTMLEncoder, which encodes HTML entities.
=item *
B<pre_tag> - deprecated.
=item *
B<post_tag> - deprecated.
=back
=head1 COPYRIGHT
Copyright 2005-2007 Marvin Humphrey
=head1 LICENSE, DISCLAIMER, BUGS, etc.
See L<KinoSearch|KinoSearch> version 0.162.
=cut
