| line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
|
1
|
|
|
|
|
|
|
package Text::Similarity::Overlaps; |
|
2
|
|
|
|
|
|
|
|
|
3
|
3
|
|
|
3
|
|
12540
|
use 5.006; |
|
|
3
|
|
|
|
|
10
|
|
|
|
3
|
|
|
|
|
109
|
|
|
4
|
3
|
|
|
3
|
|
14
|
use strict; |
|
|
3
|
|
|
|
|
4
|
|
|
|
3
|
|
|
|
|
84
|
|
|
5
|
3
|
|
|
3
|
|
13
|
use warnings; |
|
|
3
|
|
|
|
|
4
|
|
|
|
3
|
|
|
|
|
69
|
|
|
6
|
|
|
|
|
|
|
|
|
7
|
3
|
|
|
3
|
|
18
|
use Text::Similarity; |
|
|
3
|
|
|
|
|
17
|
|
|
|
3
|
|
|
|
|
128
|
|
|
8
|
3
|
|
|
3
|
|
1502
|
use Text::OverlapFinder; |
|
|
3
|
|
|
|
|
7
|
|
|
|
3
|
|
|
|
|
332
|
|
|
9
|
|
|
|
|
|
|
|
|
10
|
|
|
|
|
|
|
our @ISA = qw(Text::Similarity); |
|
11
|
|
|
|
|
|
|
|
|
12
|
|
|
|
|
|
|
# Items to export into callers namespace by default. Note: do not export |
|
13
|
|
|
|
|
|
|
# names by default without a very good reason. Use EXPORT_OK instead. |
|
14
|
|
|
|
|
|
|
# Do not simply export all your public functions/methods/constants. |
|
15
|
|
|
|
|
|
|
|
|
16
|
|
|
|
|
|
|
# This allows declaration use Text::Similarity::Overlaps ':all'; |
|
17
|
|
|
|
|
|
|
# If you do not need this, moving things directly into @EXPORT or @EXPORT_OK |
|
18
|
|
|
|
|
|
|
# will save memory. |
|
19
|
|
|
|
|
|
|
|
|
20
|
|
|
|
|
|
|
our %EXPORT_TAGS = ( 'all' => [ qw() ] ); |
|
21
|
|
|
|
|
|
|
our @EXPORT_OK = ( @{ $EXPORT_TAGS{'all'} } ); |
|
22
|
|
|
|
|
|
|
|
|
23
|
|
|
|
|
|
|
our @EXPORT = qw(); |
|
24
|
|
|
|
|
|
|
|
|
25
|
|
|
|
|
|
|
our $VERSION = '0.05'; |
|
26
|
|
|
|
|
|
|
|
|
27
|
|
|
|
|
|
|
# information about granularity is not used now |
|
28
|
|
|
|
|
|
|
# the would be that you could figure out similarity |
|
29
|
|
|
|
|
|
|
# unit by unit, that is sentence by sentence, |
|
30
|
|
|
|
|
|
|
# paragraph by paragraph, etc. however, at this |
|
31
|
|
|
|
|
|
|
# point similarity is only computed document by |
|
32
|
|
|
|
|
|
|
# document |
|
33
|
|
|
|
|
|
|
|
|
34
|
3
|
|
|
3
|
|
20
|
use constant WORD => 0; |
|
|
3
|
|
|
|
|
4
|
|
|
|
3
|
|
|
|
|
144
|
|
|
35
|
3
|
|
|
3
|
|
17
|
use constant SENTENCE => 1; |
|
|
3
|
|
|
|
|
5
|
|
|
|
3
|
|
|
|
|
137
|
|
|
36
|
3
|
|
|
3
|
|
16
|
use constant PARAGRAPH => 2; |
|
|
3
|
|
|
|
|
5
|
|
|
|
3
|
|
|
|
|
126
|
|
|
37
|
3
|
|
|
3
|
|
15
|
use constant DOCUMENT => 3; |
|
|
3
|
|
|
|
|
5
|
|
|
|
3
|
|
|
|
|
4084
|
|
|
38
|
|
|
|
|
|
|
|
|
39
|
|
|
|
|
|
|
my %finder; |
|
40
|
|
|
|
|
|
|
|
|
41
|
|
|
|
|
|
|
sub new |
|
42
|
|
|
|
|
|
|
{ |
|
43
|
4
|
|
|
4
|
0
|
472
|
my $class = shift; |
|
44
|
4
|
|
|
|
|
44
|
my $self = $class->SUPER::new (@_); |
|
45
|
4
|
50
|
|
|
|
29
|
if ($self->stoplist) { |
|
46
|
0
|
|
|
|
|
0
|
$finder{$self} = Text::OverlapFinder->new(stoplist => $self->stoplist); |
|
47
|
|
|
|
|
|
|
} |
|
48
|
|
|
|
|
|
|
else { |
|
49
|
4
|
|
|
|
|
31
|
$finder{$self} = Text::OverlapFinder->new; |
|
50
|
|
|
|
|
|
|
} |
|
51
|
|
|
|
|
|
|
|
|
52
|
|
|
|
|
|
|
#string_compare_initialize (0, undef); |
|
53
|
4
|
|
|
|
|
24
|
return $self; |
|
54
|
|
|
|
|
|
|
} |
|
55
|
|
|
|
|
|
|
|
|
56
|
|
|
|
|
|
|
|
|
57
|
|
|
|
|
|
|
sub finder : lvalue |
|
58
|
|
|
|
|
|
|
{ |
|
59
|
42
|
|
|
42
|
0
|
52
|
my $self = shift; |
|
60
|
42
|
|
|
|
|
252
|
$finder{$self}; |
|
61
|
|
|
|
|
|
|
} |
|
62
|
|
|
|
|
|
|
|
|
63
|
|
|
|
|
|
|
sub DESTROY |
|
64
|
|
|
|
|
|
|
{ |
|
65
|
4
|
|
|
4
|
|
2156
|
my $self = shift; |
|
66
|
4
|
|
|
|
|
31
|
delete $finder{$self}; |
|
67
|
|
|
|
|
|
|
} |
|
68
|
|
|
|
|
|
|
|
|
69
|
|
|
|
|
|
|
# this method requires that the input be provided in files. |
|
70
|
|
|
|
|
|
|
# this is now just a front end to getSimilarityStrings that |
|
71
|
|
|
|
|
|
|
# does file handling. Actual similarity measurements are |
|
72
|
|
|
|
|
|
|
# performed between strings in getSimilarityStrings. |
|
73
|
|
|
|
|
|
|
|
|
74
|
|
|
|
|
|
|
# this seems like it might want to move up into Similarity.pm |
|
75
|
|
|
|
|
|
|
|
|
76
|
|
|
|
|
|
|
sub getSimilarity |
|
77
|
|
|
|
|
|
|
{ |
|
78
|
28
|
|
|
28
|
0
|
29462
|
my $self = shift; |
|
79
|
|
|
|
|
|
|
|
|
80
|
|
|
|
|
|
|
# we created a separate method for getSimilarityStrings since overloading |
|
81
|
|
|
|
|
|
|
# to accept both strings and file names as input parameters to |
|
82
|
|
|
|
|
|
|
# getSimilarity would have required that we treat any file name that does |
|
83
|
|
|
|
|
|
|
# not have a corresponding file to be treated as a string, thus making it |
|
84
|
|
|
|
|
|
|
# impossible to really deal with missing file errors, and probably resulting |
|
85
|
|
|
|
|
|
|
# in quite a bit of user annoyance as 'textt1.txt' is measured for similarity |
|
86
|
|
|
|
|
|
|
# with the contents of 'text2.txt' |
|
87
|
|
|
|
|
|
|
|
|
88
|
28
|
|
|
|
|
59
|
my $file1 = shift; |
|
89
|
28
|
|
|
|
|
35
|
my $file2 = shift; |
|
90
|
|
|
|
|
|
|
|
|
91
|
|
|
|
|
|
|
# granularity is not currently supported |
|
92
|
|
|
|
|
|
|
|
|
93
|
28
|
|
|
|
|
35
|
my $granularity = shift; |
|
94
|
28
|
50
|
|
|
|
93
|
$granularity = DOCUMENT unless defined $granularity; |
|
95
|
|
|
|
|
|
|
|
|
96
|
28
|
50
|
|
|
|
563
|
unless (-e $file1) { |
|
97
|
0
|
|
|
|
|
0
|
$self->error ("The file '$file1' does not exist"); |
|
98
|
0
|
|
|
|
|
0
|
return undef; |
|
99
|
|
|
|
|
|
|
} |
|
100
|
28
|
50
|
|
|
|
310
|
unless (-e $file2) { |
|
101
|
0
|
|
|
|
|
0
|
$self->error ("The file '$file2' does not exist"); |
|
102
|
0
|
|
|
|
|
0
|
return undef; |
|
103
|
|
|
|
|
|
|
} |
|
104
|
|
|
|
|
|
|
|
|
105
|
28
|
50
|
|
|
|
916
|
unless (open (FH1, '<', $file1)) { |
|
106
|
0
|
|
|
|
|
0
|
$self->error ("Cannot open $file1: $!"); |
|
107
|
0
|
|
|
|
|
0
|
return undef; |
|
108
|
|
|
|
|
|
|
} |
|
109
|
28
|
50
|
|
|
|
1141
|
unless (open (FH2, '<', $file2)) { |
|
110
|
0
|
|
|
|
|
0
|
$self->error ("Cannot open $file2: $!"); |
|
111
|
0
|
|
|
|
|
0
|
return undef; |
|
112
|
|
|
|
|
|
|
} |
|
113
|
|
|
|
|
|
|
|
|
114
|
28
|
|
|
|
|
38
|
my $str1; |
|
115
|
|
|
|
|
|
|
my $str2; |
|
116
|
28
|
|
|
|
|
471
|
while () { |
|
117
|
100
|
|
|
|
|
278
|
$str1 .= $self->sanitizeString ($_); |
|
118
|
|
|
|
|
|
|
} |
|
119
|
28
|
|
|
|
|
538
|
$str1 =~ s/\s+/ /g; |
|
120
|
28
|
|
|
|
|
519
|
while () { |
|
121
|
66
|
|
|
|
|
159
|
$str2 .= $self->sanitizeString ($_); |
|
122
|
|
|
|
|
|
|
} |
|
123
|
28
|
|
|
|
|
597
|
$str2 =~ s/\s+/ /g; |
|
124
|
|
|
|
|
|
|
|
|
125
|
|
|
|
|
|
|
# find compounds, not working right now |
|
126
|
|
|
|
|
|
|
|
|
127
|
28
|
|
|
|
|
102
|
$str1 = $self->compoundify ($str1); |
|
128
|
28
|
|
|
|
|
72
|
$str2 = $self->compoundify ($str2); |
|
129
|
|
|
|
|
|
|
|
|
130
|
28
|
|
|
|
|
403
|
close FH1; |
|
131
|
28
|
|
|
|
|
236
|
close FH2; |
|
132
|
|
|
|
|
|
|
|
|
133
|
|
|
|
|
|
|
# add this call, to expose this method to users too (in case they want to |
|
134
|
|
|
|
|
|
|
# just measure the similarity of two strings. So we have our files converted |
|
135
|
|
|
|
|
|
|
# into strings, and now measure their similarity. |
|
136
|
|
|
|
|
|
|
|
|
137
|
28
|
|
|
|
|
91
|
my ($score, %allScores) = $self -> getSimilarityStrings ($str1,$str2); |
|
138
|
|
|
|
|
|
|
|
|
139
|
|
|
|
|
|
|
# end getSimilarity here, making sure to return similarity value from |
|
140
|
|
|
|
|
|
|
# get SimilarityStrings |
|
141
|
|
|
|
|
|
|
|
|
142
|
28
|
50
|
|
|
|
215
|
return wantarray ? ($score, %allScores) : $score; |
|
143
|
|
|
|
|
|
|
} |
|
144
|
|
|
|
|
|
|
|
|
145
|
|
|
|
|
|
|
# this method measures the similarity between two strings. If a string is empty |
|
146
|
|
|
|
|
|
|
# or missing then we generate an error and return undefined |
|
147
|
|
|
|
|
|
|
|
|
148
|
|
|
|
|
|
|
sub getSimilarityStrings { |
|
149
|
|
|
|
|
|
|
|
|
150
|
44
|
|
|
44
|
0
|
10345
|
my $self = shift; |
|
151
|
|
|
|
|
|
|
|
|
152
|
44
|
|
|
|
|
66
|
my $input1 = shift; |
|
153
|
44
|
|
|
|
|
51
|
my $input2 = shift; |
|
154
|
|
|
|
|
|
|
|
|
155
|
|
|
|
|
|
|
# check to make sure you have a string! empty file or string should be rejected |
|
156
|
|
|
|
|
|
|
|
|
157
|
44
|
100
|
|
|
|
114
|
if (!defined($input1)) { |
|
158
|
2
|
|
|
|
|
19
|
$self->error ("first input string is undefined: $!"); |
|
159
|
2
|
|
|
|
|
6
|
return undef; |
|
160
|
|
|
|
|
|
|
} |
|
161
|
42
|
50
|
|
|
|
79
|
if (!defined($input2)) { |
|
162
|
0
|
|
|
|
|
0
|
$self->error ("second input string is undefined: $!"); |
|
163
|
0
|
|
|
|
|
0
|
return undef; |
|
164
|
|
|
|
|
|
|
} |
|
165
|
|
|
|
|
|
|
|
|
166
|
|
|
|
|
|
|
# clean the strings, maybe make this optional |
|
167
|
|
|
|
|
|
|
|
|
168
|
42
|
|
|
|
|
122
|
my $str1 .= $self->sanitizeString ($input1); |
|
169
|
42
|
|
|
|
|
127
|
my $str2 .= $self->sanitizeString ($input2); |
|
170
|
|
|
|
|
|
|
|
|
171
|
|
|
|
|
|
|
# find compounds, not working right now |
|
172
|
|
|
|
|
|
|
|
|
173
|
42
|
|
|
|
|
139
|
$str1 = $self->compoundify ($str1); |
|
174
|
42
|
|
|
|
|
110
|
$str2 = $self->compoundify ($str2); |
|
175
|
|
|
|
|
|
|
|
|
176
|
|
|
|
|
|
|
# this is where we find our overlaps.... |
|
177
|
|
|
|
|
|
|
# then we score them using both the standard overlaps measure |
|
178
|
|
|
|
|
|
|
# as well as based on the lesk.pm module from WordNet-Similarity |
|
179
|
|
|
|
|
|
|
|
|
180
|
42
|
|
|
|
|
114
|
my ($overlaps, $wc1, $wc2) = $self->finder->getOverlaps ($str1, $str2); |
|
181
|
|
|
|
|
|
|
|
|
182
|
42
|
|
|
|
|
71
|
my $score = 0; |
|
183
|
42
|
|
|
|
|
84
|
my $raw_lesk = 0; |
|
184
|
42
|
|
|
|
|
163
|
my %allScores = ('wc1' => $wc1, 'wc2' => $wc2); |
|
185
|
|
|
|
|
|
|
|
|
186
|
42
|
50
|
|
|
|
201
|
if ($self->verbose) { |
|
187
|
0
|
|
|
|
|
0
|
print STDERR "keys: ", scalar keys %$overlaps, "\n"; |
|
188
|
|
|
|
|
|
|
} |
|
189
|
|
|
|
|
|
|
|
|
190
|
42
|
|
|
|
|
199
|
foreach my $key (sort keys %$overlaps) { |
|
191
|
|
|
|
|
|
|
|
|
192
|
40
|
|
|
|
|
390
|
my @words = split /\s+/, $key; |
|
193
|
|
|
|
|
|
|
|
|
194
|
40
|
50
|
|
|
|
142
|
if ($self->verbose) { |
|
195
|
0
|
|
|
|
|
0
|
print STDERR "-->'$key' len(", scalar @words, ") cnt(", $overlaps->{$key}, ")\n"; |
|
196
|
|
|
|
|
|
|
} |
|
197
|
|
|
|
|
|
|
|
|
198
|
|
|
|
|
|
|
# find out how many words match, add 1 for each match |
|
199
|
|
|
|
|
|
|
|
|
200
|
|
|
|
|
|
|
#$score += scalar @words * scalar @words * ${$overlaps}{$key}; |
|
201
|
40
|
|
|
|
|
141
|
$score += scalar @words * $overlaps->{$key}; |
|
202
|
40
|
|
|
|
|
81
|
$allScores{'raw'} = $score; |
|
203
|
|
|
|
|
|
|
# find the length of the key, square it, multiply with its |
|
204
|
|
|
|
|
|
|
# value to get the lesk score for this particular match |
|
205
|
|
|
|
|
|
|
|
|
206
|
40
|
|
|
|
|
81
|
my $value = ($#words + 1) * ($#words + 1) * ${$overlaps}{$key}; |
|
|
40
|
|
|
|
|
74
|
|
|
207
|
40
|
|
|
|
|
176
|
$raw_lesk += $value; |
|
208
|
|
|
|
|
|
|
|
|
209
|
|
|
|
|
|
|
} |
|
210
|
|
|
|
|
|
|
|
|
211
|
|
|
|
|
|
|
# fix for divide by zero error, will short circuit when score is 0 |
|
212
|
|
|
|
|
|
|
# provided by cernst at esoft.com |
|
213
|
|
|
|
|
|
|
# who reported via rt.cpan.org ticket 29902 |
|
214
|
|
|
|
|
|
|
|
|
215
|
42
|
100
|
|
|
|
107
|
if ($score == 0){ |
|
216
|
10
|
|
|
|
|
18
|
$allScores{'raw'} = 0; |
|
217
|
10
|
|
|
|
|
16
|
$allScores{'precision'} = 0; |
|
218
|
10
|
|
|
|
|
19
|
$allScores{'recall'} = 0; |
|
219
|
10
|
|
|
|
|
16
|
$allScores{'F'} = 0; |
|
220
|
10
|
|
|
|
|
14
|
$allScores{'dice'} = 0; |
|
221
|
10
|
|
|
|
|
16
|
$allScores{'E'} = 0; |
|
222
|
10
|
|
|
|
|
15
|
$allScores{'cosine'} = 0; |
|
223
|
10
|
|
|
|
|
20
|
$allScores{'raw_lesk'} = 0; |
|
224
|
10
|
|
|
|
|
17
|
$allScores{'lesk'} = 0; |
|
225
|
|
|
|
|
|
|
|
|
226
|
10
|
100
|
|
|
|
92
|
return wantarray ? ($score, %allScores) : $score; |
|
227
|
|
|
|
|
|
|
} |
|
228
|
|
|
|
|
|
|
|
|
229
|
|
|
|
|
|
|
# end of fix |
|
230
|
|
|
|
|
|
|
|
|
231
|
32
|
|
|
|
|
58
|
my $prec = $score / $wc2; |
|
232
|
32
|
|
|
|
|
43
|
my $recall = $score / $wc1; |
|
233
|
32
|
|
|
|
|
78
|
my $f = 2 * $prec * $recall / ($prec + $recall); |
|
234
|
|
|
|
|
|
|
|
|
235
|
32
|
|
|
|
|
56
|
my $dice = 2 * $score / ($wc1 + $wc2) ; |
|
236
|
32
|
|
|
|
|
48
|
my $e = 1 - $f; |
|
237
|
32
|
|
|
|
|
86
|
my $cos = $score / sqrt ($wc1 * $wc2); |
|
238
|
32
|
|
|
|
|
45
|
my $lesk = $raw_lesk/ ($wc1 * $wc2); |
|
239
|
|
|
|
|
|
|
|
|
240
|
32
|
|
|
|
|
72
|
$allScores{'precision'} = $prec; |
|
241
|
32
|
|
|
|
|
48
|
$allScores{'recall'} = $recall; |
|
242
|
32
|
|
|
|
|
61
|
$allScores{'F'} = $f; |
|
243
|
32
|
|
|
|
|
50
|
$allScores{'dice'} = $dice; |
|
244
|
32
|
|
|
|
|
59
|
$allScores{'E'} = $e; |
|
245
|
32
|
|
|
|
|
54
|
$allScores{'cosine'} = $cos; |
|
246
|
32
|
|
|
|
|
49
|
$allScores{'raw_lesk'} = $raw_lesk; |
|
247
|
32
|
|
|
|
|
59
|
$allScores{'lesk'} = $lesk; |
|
248
|
|
|
|
|
|
|
|
|
249
|
|
|
|
|
|
|
# display them, if requested |
|
250
|
32
|
50
|
|
|
|
96
|
if ($self->verbose) { |
|
251
|
0
|
|
|
|
|
0
|
print STDERR "wc 1: $wc1\n"; |
|
252
|
0
|
|
|
|
|
0
|
print STDERR "wc 2: $wc2\n"; |
|
253
|
0
|
|
|
|
|
0
|
print STDERR " Raw score: $score\n"; |
|
254
|
0
|
|
|
|
|
0
|
print STDERR " Precision: $prec\n"; |
|
255
|
0
|
|
|
|
|
0
|
print STDERR " Recall : $recall\n"; |
|
256
|
0
|
|
|
|
|
0
|
print STDERR " F-measure: $f\n"; |
|
257
|
0
|
|
|
|
|
0
|
print STDERR " Dice : $dice\n"; |
|
258
|
0
|
|
|
|
|
0
|
print STDERR " E-measure: $e\n"; |
|
259
|
0
|
|
|
|
|
0
|
print STDERR " Cosine : $cos\n"; |
|
260
|
0
|
|
|
|
|
0
|
print STDERR " Raw lesk : $raw_lesk\n"; |
|
261
|
0
|
|
|
|
|
0
|
print STDERR " Lesk : $lesk\n"; |
|
262
|
|
|
|
|
|
|
|
|
263
|
|
|
|
|
|
|
# |
|
264
|
|
|
|
|
|
|
# include might be the jaccard coefficient .... |
|
265
|
|
|
|
|
|
|
# jaccard is 2*raw_score / union of string1 and string2 |
|
266
|
|
|
|
|
|
|
# |
|
267
|
|
|
|
|
|
|
} |
|
268
|
|
|
|
|
|
|
|
|
269
|
32
|
100
|
|
|
|
106
|
if ($self->normalize) { |
|
270
|
16
|
|
|
|
|
24
|
$score = $f; |
|
271
|
|
|
|
|
|
|
} |
|
272
|
|
|
|
|
|
|
|
|
273
|
32
|
100
|
|
|
|
426
|
return wantarray ? ($score, %allScores) : $score; |
|
274
|
|
|
|
|
|
|
} |
|
275
|
|
|
|
|
|
|
|
|
276
|
0
|
|
|
0
|
0
|
|
sub doStop {0} |
|
277
|
|
|
|
|
|
|
|
|
278
|
|
|
|
|
|
|
1; |
|
279
|
|
|
|
|
|
|
|
|
280
|
|
|
|
|
|
|
__END__ |