line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
package Lingua::EN::Segmenter::TextTiling; |
2
|
|
|
|
|
|
|
|
3
|
|
|
|
|
|
|
=head1 NAME |
4
|
|
|
|
|
|
|
|
5
|
|
|
|
|
|
|
Lingua::EN::Segmenter::TextTiling - Segment text using the TextTiling method |
6
|
|
|
|
|
|
|
|
7
|
|
|
|
|
|
|
=head1 SYNOPSIS |
8
|
|
|
|
|
|
|
|
9
|
|
|
|
|
|
|
use Lingua::EN::Segmenter::TextTiling qw(segments); |
10
|
|
|
|
|
|
|
use lib '.'; |
11
|
|
|
|
|
|
|
|
12
|
|
|
|
|
|
|
my $text = <
|
13
|
|
|
|
|
|
|
Lingua::EN::Segmenter is a useful module that allows text to be split up |
14
|
|
|
|
|
|
|
into words, paragraphs, segments, and tiles. |
15
|
|
|
|
|
|
|
|
16
|
|
|
|
|
|
|
Paragraphs are by default indicated by blank lines. Known segment breaks are |
17
|
|
|
|
|
|
|
indicated by a line with only the word "segment_break" in it. |
18
|
|
|
|
|
|
|
|
19
|
|
|
|
|
|
|
The module detects paragraphs that are unrelated to each other by comparing |
20
|
|
|
|
|
|
|
the number of words per-paragraph that are related. The algorithm is designed |
21
|
|
|
|
|
|
|
to work only on long segments. |
22
|
|
|
|
|
|
|
|
23
|
|
|
|
|
|
|
SOUTH OF BAGHDAD, Iraq (CNN) -- Seven U.S. troops freed Sunday after being |
24
|
|
|
|
|
|
|
held by Iraqi forces arrived by helicopter at a base south of Baghdad and were |
25
|
|
|
|
|
|
|
transferred to a C-130 transport plane headed for Kuwait, CNN's Bob Franken |
26
|
|
|
|
|
|
|
reported from the scene. |
27
|
|
|
|
|
|
|
|
28
|
|
|
|
|
|
|
EOT |
29
|
|
|
|
|
|
|
|
30
|
|
|
|
|
|
|
my $num_segment_breaks = 1; |
31
|
|
|
|
|
|
|
my @segments = segments($num_segment_breaks,$text); |
32
|
|
|
|
|
|
|
print $segments[0]; # Prints the first three paragraphs of the above text |
33
|
|
|
|
|
|
|
print "\n----------SEGMENT_BREAK----------\n"; |
34
|
|
|
|
|
|
|
print $segments[1]; # Prints the last paragraph of the above text |
35
|
|
|
|
|
|
|
|
36
|
|
|
|
|
|
|
# This module can also be used in an object-oriented fashion |
37
|
|
|
|
|
|
|
my $splitter = new Lingua::EN::Splitter; |
38
|
|
|
|
|
|
|
@words = $splitter->words($text); |
39
|
|
|
|
|
|
|
|
40
|
|
|
|
|
|
|
=head1 DESCRIPTION |
41
|
|
|
|
|
|
|
|
42
|
|
|
|
|
|
|
See synopsis. |
43
|
|
|
|
|
|
|
|
44
|
|
|
|
|
|
|
=head1 EXTENDING |
45
|
|
|
|
|
|
|
|
46
|
|
|
|
|
|
|
This module is designed to be easily extendable. Feel free to extend from this |
47
|
|
|
|
|
|
|
module when designing alternate methods for text segmentation. |
48
|
|
|
|
|
|
|
|
49
|
|
|
|
|
|
|
=head1 AUTHORS |
50
|
|
|
|
|
|
|
|
51
|
|
|
|
|
|
|
David James |
52
|
|
|
|
|
|
|
|
53
|
|
|
|
|
|
|
=head1 SEE ALSO |
54
|
|
|
|
|
|
|
|
55
|
|
|
|
|
|
|
L, L, |
56
|
|
|
|
|
|
|
L |
57
|
|
|
|
|
|
|
|
58
|
|
|
|
|
|
|
=head1 LICENSE |
59
|
|
|
|
|
|
|
|
60
|
|
|
|
|
|
|
Copyright (c) 2002 David James |
61
|
|
|
|
|
|
|
All rights reserved. |
62
|
|
|
|
|
|
|
This program is free software; you can redistribute it and/or |
63
|
|
|
|
|
|
|
modify it under the same terms as Perl itself. |
64
|
|
|
|
|
|
|
|
65
|
|
|
|
|
|
|
=cut |
66
|
|
|
|
|
|
|
|
67
|
|
|
|
|
|
|
$VERSION = 0.10; |
68
|
|
|
|
|
|
|
@EXPORT_OK = qw( |
69
|
|
|
|
|
|
|
segment |
70
|
|
|
|
|
|
|
segments |
71
|
|
|
|
|
|
|
set_tiles_per_block |
72
|
|
|
|
|
|
|
set_number_of_smoothing_rounds |
73
|
|
|
|
|
|
|
set_tokens_per_tile |
74
|
|
|
|
|
|
|
set_paragraph_regexp |
75
|
|
|
|
|
|
|
set_non_word_regexp |
76
|
|
|
|
|
|
|
set_locale |
77
|
|
|
|
|
|
|
set_stop_words |
78
|
|
|
|
|
|
|
); |
79
|
|
|
|
|
|
|
|
80
|
1
|
|
|
1
|
|
3109
|
use Math::Vector::SortIndexes qw(sort_indexes_descending); |
|
1
|
|
|
|
|
212
|
|
|
1
|
|
|
|
|
81
|
|
81
|
1
|
|
|
1
|
|
849
|
use Math::VecStat qw(average min sum); |
|
1
|
|
|
|
|
1336
|
|
|
1
|
|
|
|
|
86
|
|
82
|
1
|
|
|
1
|
|
886
|
use Math::HashSum qw(hashsum); |
|
1
|
|
|
|
|
178
|
|
|
1
|
|
|
|
|
61
|
|
83
|
|
|
|
|
|
|
|
84
|
1
|
|
|
1
|
|
7
|
use base 'Lingua::EN::Segmenter'; |
|
1
|
|
|
|
|
1
|
|
|
1
|
|
|
|
|
1476
|
|
85
|
|
|
|
|
|
|
use strict; |
86
|
|
|
|
|
|
|
|
87
|
|
|
|
|
|
|
|
88
|
|
|
|
|
|
|
# Create a new instance of the object |
89
|
|
|
|
|
|
|
sub new { |
90
|
|
|
|
|
|
|
my $class = shift; |
91
|
|
|
|
|
|
|
$class->SUPER::new( |
92
|
|
|
|
|
|
|
TILES_PER_BLOCK=>7, |
93
|
|
|
|
|
|
|
NUMBER_OF_SMOOTHING_ROUNDS=>2, |
94
|
|
|
|
|
|
|
@_ |
95
|
|
|
|
|
|
|
); |
96
|
|
|
|
|
|
|
} |
97
|
|
|
|
|
|
|
|
98
|
|
|
|
|
|
|
# Segment a piece of text |
99
|
|
|
|
|
|
|
sub segment { |
100
|
|
|
|
|
|
|
my ($self, $num_segments, $input) = @_; |
101
|
|
|
|
|
|
|
|
102
|
|
|
|
|
|
|
# Get smoothed depth scores |
103
|
|
|
|
|
|
|
my $scores = $self->smoothed_depth_scores($input); |
104
|
|
|
|
|
|
|
|
105
|
|
|
|
|
|
|
# Get paragraph breaks |
106
|
|
|
|
|
|
|
my $breaks = $self->{splitter}->paragraph_breaks($input); |
107
|
|
|
|
|
|
|
|
108
|
|
|
|
|
|
|
# Get predicted segment breaks |
109
|
|
|
|
|
|
|
return $self->tile2segment($num_segments, $breaks, $scores); |
110
|
|
|
|
|
|
|
} |
111
|
|
|
|
|
|
|
|
112
|
|
|
|
|
|
|
sub set_tiles_per_block { |
113
|
|
|
|
|
|
|
my $self = shift; |
114
|
|
|
|
|
|
|
$self->{TILES_PER_BLOCK} = shift; |
115
|
|
|
|
|
|
|
} |
116
|
|
|
|
|
|
|
|
117
|
|
|
|
|
|
|
sub set_number_of_smoothing_rounds { |
118
|
|
|
|
|
|
|
my $self = shift; |
119
|
|
|
|
|
|
|
$self->{NUMBER_OF_SMOOTHING_ROUNDS} = shift; |
120
|
|
|
|
|
|
|
} |
121
|
|
|
|
|
|
|
|
122
|
|
|
|
|
|
|
|
123
|
|
|
|
|
|
|
###################################################################### |
124
|
|
|
|
|
|
|
# PRIVATE METHODS |
125
|
|
|
|
|
|
|
###################################################################### |
126
|
|
|
|
|
|
|
|
127
|
|
|
|
|
|
|
# Accept as input the scores of the tiles. Output segment scores. |
128
|
|
|
|
|
|
|
sub tile2segment { |
129
|
|
|
|
|
|
|
my ($self, $num_segments, $breaks, $scores) = @_; |
130
|
|
|
|
|
|
|
|
131
|
|
|
|
|
|
|
my @indexes = sort_indexes_descending @$scores; |
132
|
|
|
|
|
|
|
|
133
|
|
|
|
|
|
|
my @too_close = ( |
134
|
|
|
|
|
|
|
-$self->{MIN_SEGMENT_SIZE}..-1, 1..$self->{MIN_SEGMENT_SIZE} |
135
|
|
|
|
|
|
|
); |
136
|
|
|
|
|
|
|
my @direction = qw(L R); |
137
|
|
|
|
|
|
|
my (%segments, %verbose, $cut_off_depth); |
138
|
|
|
|
|
|
|
|
139
|
|
|
|
|
|
|
# Calculate the most likely segment breaks |
140
|
|
|
|
|
|
|
GAP: foreach my $i (@indexes) { |
141
|
|
|
|
|
|
|
my $tile_no = $i + $self->{TILES_PER_BLOCK}; |
142
|
|
|
|
|
|
|
my $closest_break = (min( map { abs($_ - $tile_no) } @$breaks ))[1]; |
143
|
|
|
|
|
|
|
$segments{$closest_break+$_} and next GAP for (@too_close); |
144
|
|
|
|
|
|
|
$segments{$closest_break} .= |
145
|
|
|
|
|
|
|
$direction[$breaks->[$closest_break] > $tile_no]; |
146
|
|
|
|
|
|
|
$verbose{$tile_no} = [ $tile_no, $scores->[$i], $closest_break ]; |
147
|
|
|
|
|
|
|
$cut_off_depth = $scores->[$i]; |
148
|
|
|
|
|
|
|
last if keys %segments == $num_segments; |
149
|
|
|
|
|
|
|
} |
150
|
|
|
|
|
|
|
|
151
|
|
|
|
|
|
|
# Verbose output |
152
|
|
|
|
|
|
|
if ($self->{VERBOSE}) { |
153
|
|
|
|
|
|
|
printf "Cut-off depth = %6.4f\n\n", $cut_off_depth; |
154
|
|
|
|
|
|
|
print " Gap Depth Para\n"; |
155
|
|
|
|
|
|
|
foreach (sort { $a <=> $b } keys %verbose) { |
156
|
|
|
|
|
|
|
printf "%4d %6.3f %4d\n", @{$verbose{$_}} |
157
|
|
|
|
|
|
|
} |
158
|
|
|
|
|
|
|
print "\n"; |
159
|
|
|
|
|
|
|
} |
160
|
|
|
|
|
|
|
return \%segments; |
161
|
|
|
|
|
|
|
} |
162
|
|
|
|
|
|
|
|
163
|
|
|
|
|
|
|
# Calculate depth scores based on a list of gap scores |
164
|
|
|
|
|
|
|
sub depth { |
165
|
|
|
|
|
|
|
no warnings; |
166
|
|
|
|
|
|
|
|
167
|
|
|
|
|
|
|
my $self = shift; |
168
|
|
|
|
|
|
|
my @score = @{$_[0]}; |
169
|
|
|
|
|
|
|
my @depth; |
170
|
|
|
|
|
|
|
for my $i (1..$#score) { |
171
|
|
|
|
|
|
|
$depth[$i] = $score[$i-1] + $score[$i+1] - 2*$score[$i]; |
172
|
|
|
|
|
|
|
} |
173
|
|
|
|
|
|
|
$depth[0] = $score[1] - $score[0]; |
174
|
|
|
|
|
|
|
$depth[$#score] = $score[-2] - $score[-1]; |
175
|
|
|
|
|
|
|
return \@depth; |
176
|
|
|
|
|
|
|
} |
177
|
|
|
|
|
|
|
|
178
|
|
|
|
|
|
|
# Given some depth scores, smooth them. |
179
|
|
|
|
|
|
|
sub smooth { |
180
|
|
|
|
|
|
|
my $self = shift; |
181
|
|
|
|
|
|
|
my @depth = @{$_[0]}; |
182
|
|
|
|
|
|
|
unshift @depth, $depth[0]; |
183
|
|
|
|
|
|
|
push @depth, $depth[-1]; |
184
|
|
|
|
|
|
|
for (1..$self->{NUMBER_OF_SMOOTHING_ROUNDS}) { |
185
|
|
|
|
|
|
|
foreach my $j (1..$#depth-1) { |
186
|
|
|
|
|
|
|
$depth[$j] = average $depth[$j-1], $depth[$j], $depth[$j+1]; |
187
|
|
|
|
|
|
|
} |
188
|
|
|
|
|
|
|
} |
189
|
|
|
|
|
|
|
return [ @depth[1..$#depth-1] ]; |
190
|
|
|
|
|
|
|
} |
191
|
|
|
|
|
|
|
|
192
|
|
|
|
|
|
|
# Take text as input and output a list of smoothed depth scores |
193
|
|
|
|
|
|
|
sub smoothed_depth_scores { |
194
|
|
|
|
|
|
|
my ($self, $input) = @_; |
195
|
|
|
|
|
|
|
my $words = $self->{splitter}->words($input); |
196
|
|
|
|
|
|
|
my $tiles = $self->{splitter}->tile($words); |
197
|
|
|
|
|
|
|
my $depth_scores = $self->depth($self->gap_scores($tiles)); |
198
|
|
|
|
|
|
|
return $self->smooth($depth_scores); |
199
|
|
|
|
|
|
|
} |
200
|
|
|
|
|
|
|
|
201
|
|
|
|
|
|
|
# Scores for the gap between two tiles |
202
|
|
|
|
|
|
|
sub gap_scores { |
203
|
|
|
|
|
|
|
my $self = shift; |
204
|
|
|
|
|
|
|
my @tiles = @{$_[0]}; |
205
|
|
|
|
|
|
|
my $TILES_PER_BLOCK = $self->{TILES_PER_BLOCK}; |
206
|
|
|
|
|
|
|
my (@score, $i); |
207
|
|
|
|
|
|
|
for $i ($TILES_PER_BLOCK .. @tiles-$TILES_PER_BLOCK) { |
208
|
|
|
|
|
|
|
my $L = $i-$TILES_PER_BLOCK; |
209
|
|
|
|
|
|
|
my $R = $i+$TILES_PER_BLOCK-1; |
210
|
|
|
|
|
|
|
my %l = hashsum map { %$_ } @tiles[$L..$i-1]; |
211
|
|
|
|
|
|
|
my %r = hashsum map { %$_ } @tiles[$i..$R]; |
212
|
|
|
|
|
|
|
my %all = map { %$_ } @tiles[$L..$R]; |
213
|
|
|
|
|
|
|
my $numerator = sum map { $l{$_}*$r{$_} } keys %all; |
214
|
|
|
|
|
|
|
my $denom1 = sum map { $l{$_}*$l{$_} } keys %all; |
215
|
|
|
|
|
|
|
my $denom2 = sum map { $r{$_}*$r{$_} } keys %all; |
216
|
|
|
|
|
|
|
push @score, $numerator/sqrt($denom1*$denom2); |
217
|
|
|
|
|
|
|
} |
218
|
|
|
|
|
|
|
return \@score; |
219
|
|
|
|
|
|
|
} |
220
|
|
|
|
|
|
|
|
221
|
|
|
|
|
|
|
|
222
|
|
|
|
|
|
|
1; |
223
|
|
|
|
|
|
|
|
224
|
|
|
|
|
|
|
|
225
|
|
|
|
|
|
|
|
226
|
|
|
|
|
|
|
|
227
|
|
|
|
|
|
|
|