excite 2.1.1
Sign up to get free protection for your applications and to get access to all the features.
- data/.gitignore +11 -0
- data/.rspec +1 -0
- data/Gemfile +8 -0
- data/Gemfile.lock +69 -0
- data/LICENSE +22 -0
- data/README.md +46 -0
- data/Rakefile +24 -0
- data/config/citation_cleanup_rules.yml +68 -0
- data/config/parscit_features.yml +55 -0
- data/excite.gemspec +30 -0
- data/lib/excite/array_helpers.rb +27 -0
- data/lib/excite/citation.rb +48 -0
- data/lib/excite/crfparser.rb +322 -0
- data/lib/excite/postprocessor.rb +252 -0
- data/lib/excite/preprocessor.rb +107 -0
- data/lib/excite/resources/dicts/female-names +4954 -0
- data/lib/excite/resources/dicts/first-names +27926 -0
- data/lib/excite/resources/dicts/male-names +3901 -0
- data/lib/excite/resources/dicts/months +24 -0
- data/lib/excite/resources/dicts/places +43109 -0
- data/lib/excite/resources/dicts/publishers +654 -0
- data/lib/excite/resources/dicts/surnames +146259 -0
- data/lib/excite/resources/html.template +84 -0
- data/lib/excite/resources/html_model +0 -0
- data/lib/excite/resources/model +0 -0
- data/lib/excite/resources/parsCit.template +76 -0
- data/lib/excite/resources/trainingdata/tagged_html_references.txt +500 -0
- data/lib/excite/resources/trainingdata/tagged_references.txt +500 -0
- data/lib/excite/resources/trainingdata/verify.rb +97 -0
- data/lib/excite/token_features.rb +313 -0
- data/lib/excite/version.rb +7 -0
- data/lib/excite.rb +13 -0
- data/model/test/analysis.csv +54 -0
- data/model/test/array_helpers.rb +30 -0
- data/model/test/html-analysis.csv +60 -0
- data/model/test/html-output.txt +19893 -0
- data/model/test/model_test.rb +306 -0
- data/model/test/output.txt +16742 -0
- data/spec/excite/citation_spec.rb +128 -0
- data/spec/excite/crfparser_spec.rb +118 -0
- data/spec/excite/postprocessor_spec.rb +68 -0
- data/spec/excite/token_features_spec.rb +641 -0
- data/spec/spec_helper.rb +4 -0
- metadata +222 -0
@@ -0,0 +1,322 @@
|
|
1
|
+
# encoding: UTF-8
|
2
|
+
|
3
|
+
require 'tempfile'
|
4
|
+
require 'nokogiri'
|
5
|
+
require 'cgi'
|
6
|
+
require 'engtagger'
|
7
|
+
|
8
|
+
module Excite
|
9
|
+
|
10
|
+
class CRFParser
|
11
|
+
|
12
|
+
attr_reader :feature_order
|
13
|
+
attr_reader :token_features
|
14
|
+
|
15
|
+
include TokenFeatures
|
16
|
+
include Preprocessor
|
17
|
+
include Postprocessor
|
18
|
+
|
19
|
+
DIR = File.dirname(__FILE__)
|
20
|
+
TAGGED_REFERENCES = "#{DIR}/resources/trainingdata/tagged_references.txt"
|
21
|
+
TAGGED_HTML_REFERENCES = "#{DIR}/resources/trainingdata/tagged_html_references.txt"
|
22
|
+
TRAINING_DATA = "#{DIR}/resources/trainingdata/training_data.txt"
|
23
|
+
MODEL_FILE = "#{DIR}/resources/model"
|
24
|
+
HTML_MODEL_FILE = "#{DIR}/resources/html_model"
|
25
|
+
TEMPLATE_FILE = "#{DIR}/resources/parsCit.template"
|
26
|
+
HTML_TEMPLATE_FILE = "#{DIR}/resources/html.template"
|
27
|
+
CONFIG_FILE = "#{DIR}/../../config/parscit_features.yml"
|
28
|
+
|
29
|
+
# Feature functions must be performed in alphabetical order, since
|
30
|
+
# later functions may depend on earlier ones.
|
31
|
+
# TODO This seems pretty confusing and dependent on the current features.
|
32
|
+
def initialize(mode=:string)
|
33
|
+
@mode = mode
|
34
|
+
|
35
|
+
f = File.open(CONFIG_FILE, 'r')
|
36
|
+
hsh = YAML::load(f)[mode.to_s]
|
37
|
+
@feature_order = hsh["feature_order"].map(&:to_sym)
|
38
|
+
@token_features = hsh["feature_order"].sort.map(&:to_sym)
|
39
|
+
end
|
40
|
+
|
41
|
+
def model
|
42
|
+
@model ||= CRFPP::Tagger.new("-m #{default_model_file} -v 1");
|
43
|
+
end
|
44
|
+
|
45
|
+
def parse(str, presumed_author=nil)
|
46
|
+
raw_string = str.dup
|
47
|
+
|
48
|
+
toks, features = str_2_features(str, false, presumed_author)
|
49
|
+
tags, overall_prob, tag_probs = eval_crfpp(features, model)
|
50
|
+
|
51
|
+
ret = {}
|
52
|
+
tags.each_with_index { |t, i| (ret[t] ||= []) << toks[i].for_join(toks[i-1]) }
|
53
|
+
ret.each { |k, v| ret[k] = v.join('').strip }
|
54
|
+
|
55
|
+
normalize_fields(ret)
|
56
|
+
ret['raw_string'] = raw_string
|
57
|
+
[ret, overall_prob, tag_probs]
|
58
|
+
end
|
59
|
+
|
60
|
+
def eval_crfpp(feat_seq, model)
|
61
|
+
model.clear
|
62
|
+
feat_seq.each {|vec|
|
63
|
+
line = vec.join(" ").strip
|
64
|
+
raise unless model.add(line)
|
65
|
+
}
|
66
|
+
raise unless model.parse
|
67
|
+
tags = []
|
68
|
+
probs = {}
|
69
|
+
feat_seq.length.times {|i|
|
70
|
+
tags << model.y2(i)
|
71
|
+
probs[model.y2(i)] ||= 1
|
72
|
+
probs[model.y2(i)] *= model.prob(i)
|
73
|
+
}
|
74
|
+
[tags, model.prob, probs]
|
75
|
+
end
|
76
|
+
|
77
|
+
def self.strip_punct(str)
|
78
|
+
toknp = str.gsub(/[^\w]/, '')
|
79
|
+
toknp = "EMPTY" if toknp.blank? # TODO Seems maybe hacky
|
80
|
+
toknp
|
81
|
+
end
|
82
|
+
|
83
|
+
def normalize_input_author(str)
|
84
|
+
return nil if str.blank?
|
85
|
+
str.split.map(&:downcase).map{ |t| self.class.strip_punct(t) }.select{ |s| s.length > 2 }
|
86
|
+
end
|
87
|
+
|
88
|
+
def prepare_token_data(raw_string, training=false)
|
89
|
+
if training
|
90
|
+
tags = tagged_string_2_tags(raw_string.strip)
|
91
|
+
|
92
|
+
labels, raw_string, joined_tokens = [], '', ''
|
93
|
+
tags.each do |tag|
|
94
|
+
raw = CGI.unescapeHTML(tag.inner_html)
|
95
|
+
|
96
|
+
label = tag.name
|
97
|
+
raise "Invalid label #{label} for:\n#{str}" if label.present? && !recognized_labels.include?(label)
|
98
|
+
|
99
|
+
toks = str_2_tokens(raw)
|
100
|
+
|
101
|
+
labels << [label, joined_tokens.length]
|
102
|
+
joined_tokens += toks.map(&:raw).join
|
103
|
+
raw_string += "\n#{raw}"
|
104
|
+
end
|
105
|
+
end
|
106
|
+
|
107
|
+
tokens = str_2_tokens(raw_string.strip)
|
108
|
+
|
109
|
+
if training
|
110
|
+
joined_tokens = ''
|
111
|
+
label, _ = labels.shift
|
112
|
+
next_label, end_idx = labels.shift unless labels.empty?
|
113
|
+
|
114
|
+
tokens.each do |tok|
|
115
|
+
tok.label = label
|
116
|
+
joined_tokens += tok.raw
|
117
|
+
if joined_tokens.length == end_idx
|
118
|
+
label = next_label
|
119
|
+
next_label, end_idx = labels.shift unless labels.empty?
|
120
|
+
elsif joined_tokens.length > end_idx && !labels.empty?
|
121
|
+
raise "Tokens do not match labels"
|
122
|
+
end
|
123
|
+
end
|
124
|
+
raise "Unused label" unless labels.empty?
|
125
|
+
end
|
126
|
+
|
127
|
+
self.clear
|
128
|
+
|
129
|
+
return tokens
|
130
|
+
end
|
131
|
+
|
132
|
+
def tagger
|
133
|
+
@tagger ||= EngTagger.new
|
134
|
+
end
|
135
|
+
|
136
|
+
def tagged_string_2_tags(str)
|
137
|
+
str = "<string>#{str}</string>"
|
138
|
+
node = Nokogiri::XML.fragment(str).css('string')
|
139
|
+
node.children.reject(&:text?)
|
140
|
+
end
|
141
|
+
|
142
|
+
def str_2_tokens(str)
|
143
|
+
if @mode == :html
|
144
|
+
toks = html_str_2_tokens(str)
|
145
|
+
elsif @mode == :string
|
146
|
+
toks = text_str_2_tokens(str)
|
147
|
+
end
|
148
|
+
|
149
|
+
toks.reject { |t| t.empty? }
|
150
|
+
end
|
151
|
+
|
152
|
+
def recognized_labels
|
153
|
+
if @mode == :string
|
154
|
+
["author", "title", "editor", "booktitle", "date", "journal", "volume", "institution", "pages", "location", "publisher", "note", "tech"]
|
155
|
+
elsif @mode == :html
|
156
|
+
["author", "title", "editor", "booktitle", "date", "journal", "volume", "institution", "pages", "location", "publisher", "note", "workid", "link", "bullet"]
|
157
|
+
else
|
158
|
+
[]
|
159
|
+
end
|
160
|
+
end
|
161
|
+
|
162
|
+
def html_str_2_tokens(str)
|
163
|
+
html = Nokogiri::HTML.fragment(str.gsub('>', '> ')) # gsub to ensure strings in separate tags are always separate tokens even if HTML is bad
|
164
|
+
|
165
|
+
tokens = []
|
166
|
+
html.traverse do |node|
|
167
|
+
tokens += html_text_node_2_tokens(node) if node.text?
|
168
|
+
end
|
169
|
+
tokens
|
170
|
+
end
|
171
|
+
|
172
|
+
def html_text_node_2_tokens(node)
|
173
|
+
text = CGI.unescapeHTML(node.text)
|
174
|
+
return [] if text.blank?
|
175
|
+
|
176
|
+
tokens = text_str_2_tokens(text)
|
177
|
+
tokens.each_with_index { |tok, i| tok.is_in_node!(node, i, tokens.length) }
|
178
|
+
tokens
|
179
|
+
end
|
180
|
+
|
181
|
+
def text_str_2_tokens(text)
|
182
|
+
tagged = tagger.add_tags(normalize_citation(text))
|
183
|
+
tags = tagged_string_2_tags(tagged.gsub('&','&')) # EngTagger has legitimately added angle brackets which are meaningful in XML, but angle-brackets predate EngTagger and are semantic
|
184
|
+
tags.map { |tag| Token.new(tag.text, tag.name) }
|
185
|
+
end
|
186
|
+
|
187
|
+
# calculate features on the full citation string
|
188
|
+
def str_2_features(raw_string, training=false, presumed_author=nil)
|
189
|
+
features = []
|
190
|
+
tokens = prepare_token_data(raw_string, training)
|
191
|
+
|
192
|
+
author_names = normalize_input_author(presumed_author)
|
193
|
+
|
194
|
+
tokens.each_with_index do |tok, toki|
|
195
|
+
raise "All tokens must be labeled" if training && tok.label.nil?
|
196
|
+
|
197
|
+
feats = {}
|
198
|
+
|
199
|
+
@token_features.each {|f|
|
200
|
+
feats[f] = self.send(f, tokens, toki, author_names)
|
201
|
+
}
|
202
|
+
|
203
|
+
features << [tok.raw]
|
204
|
+
@feature_order.each {|f| features.last << feats[f]}
|
205
|
+
features.last << tok.label if training
|
206
|
+
end
|
207
|
+
|
208
|
+
[tokens, features]
|
209
|
+
end
|
210
|
+
|
211
|
+
def write_training_file(tagged_refs=nil, training_data=TRAINING_DATA)
|
212
|
+
tagged_refs ||= default_tagged_references
|
213
|
+
|
214
|
+
fin = File.open(tagged_refs, 'r')
|
215
|
+
fout = File.open(training_data, 'w')
|
216
|
+
x = 0
|
217
|
+
while l = fin.gets
|
218
|
+
_, data = str_2_features(l.strip, true)
|
219
|
+
data.each {|line| fout.write("#{line.join(" ")}\n") }
|
220
|
+
fout.write("\n")
|
221
|
+
end
|
222
|
+
|
223
|
+
fin.close
|
224
|
+
fout.flush
|
225
|
+
fout.close
|
226
|
+
end
|
227
|
+
|
228
|
+
def train(tagged_refs=nil, model=nil, template=nil, training_data=nil)
|
229
|
+
tagged_refs ||= default_tagged_references
|
230
|
+
model ||= default_model_file
|
231
|
+
template ||= default_template_file
|
232
|
+
|
233
|
+
if training_data.nil?
|
234
|
+
training_data = TRAINING_DATA
|
235
|
+
write_training_file(tagged_refs, training_data)
|
236
|
+
end
|
237
|
+
|
238
|
+
`crf_learn #{template} #{training_data} #{model} -f3 1>&2`
|
239
|
+
end
|
240
|
+
|
241
|
+
def default_tagged_references
|
242
|
+
if @mode == :string
|
243
|
+
TAGGED_REFERENCES
|
244
|
+
elsif @mode == :html
|
245
|
+
TAGGED_HTML_REFERENCES
|
246
|
+
else
|
247
|
+
raise "Unknown mode: #{@mode}"
|
248
|
+
end
|
249
|
+
end
|
250
|
+
|
251
|
+
def default_model_file
|
252
|
+
if @mode == :string
|
253
|
+
MODEL_FILE
|
254
|
+
elsif @mode == :html
|
255
|
+
HTML_MODEL_FILE
|
256
|
+
else
|
257
|
+
raise "Unknown mode: #{@mode}"
|
258
|
+
end
|
259
|
+
end
|
260
|
+
|
261
|
+
def default_template_file
|
262
|
+
if @mode == :string
|
263
|
+
TEMPLATE_FILE
|
264
|
+
elsif @mode == :html
|
265
|
+
HTML_TEMPLATE_FILE
|
266
|
+
else
|
267
|
+
raise "Unknown mode: #{@mode}"
|
268
|
+
end
|
269
|
+
end
|
270
|
+
|
271
|
+
end
|
272
|
+
|
273
|
+
class TrainingError < Exception; end
|
274
|
+
|
275
|
+
class Token
|
276
|
+
|
277
|
+
attr_reader :node, :idx_in_node, :node_token_count, :part_of_speech
|
278
|
+
attr_accessor :label
|
279
|
+
|
280
|
+
def initialize(str, part_of_speech=nil)
|
281
|
+
@str = str
|
282
|
+
@part_of_speech = part_of_speech
|
283
|
+
end
|
284
|
+
|
285
|
+
def is_in_node!(node, idx_in_node, node_token_count)
|
286
|
+
@node = node
|
287
|
+
@idx_in_node = idx_in_node
|
288
|
+
@node_token_count = node_token_count
|
289
|
+
end
|
290
|
+
|
291
|
+
def raw
|
292
|
+
@str
|
293
|
+
end
|
294
|
+
|
295
|
+
def np
|
296
|
+
@np ||= CRFParser.strip_punct(@str)
|
297
|
+
end
|
298
|
+
|
299
|
+
def lcnp
|
300
|
+
@lcnp ||= np == "EMPTY" ? np : np.downcase
|
301
|
+
end
|
302
|
+
|
303
|
+
def empty?
|
304
|
+
raw.strip.blank?
|
305
|
+
end
|
306
|
+
|
307
|
+
def to_s
|
308
|
+
"{#{raw}}"
|
309
|
+
end
|
310
|
+
|
311
|
+
def for_join(prev)
|
312
|
+
if ['pp','ppc','ppr','pps','rrb', 'pos'].include?(part_of_speech)
|
313
|
+
raw
|
314
|
+
elsif prev && ['ppd','ppl','lrb'].include?(prev.part_of_speech)
|
315
|
+
raw
|
316
|
+
else
|
317
|
+
" "+raw
|
318
|
+
end
|
319
|
+
end
|
320
|
+
end
|
321
|
+
|
322
|
+
end
|
@@ -0,0 +1,252 @@
|
|
1
|
+
# encoding: UTF-8
|
2
|
+
|
3
|
+
module Excite
|
4
|
+
|
5
|
+
module Postprocessor
|
6
|
+
|
7
|
+
def normalize_fields(citation_hsh)
|
8
|
+
citation_hsh.keys.each {|key| self.send("normalize_#{key}", citation_hsh) }
|
9
|
+
citation_hsh
|
10
|
+
end
|
11
|
+
|
12
|
+
def method_missing(m, *args, &block)
|
13
|
+
# Call normalize on any fields that don't have their own normalization
|
14
|
+
# method defined
|
15
|
+
if m.to_s =~ /^normalize/
|
16
|
+
m.to_s =~ /normalize_(.*)$/
|
17
|
+
normalize($1, *args)
|
18
|
+
else super
|
19
|
+
end
|
20
|
+
end
|
21
|
+
|
22
|
+
# default normalization function for all fields that do not have their
|
23
|
+
# own normalization
|
24
|
+
# Strip any leading and/or trailing punctuation and space
|
25
|
+
def normalize(key, hsh)
|
26
|
+
hsh[key].gsub!(/^[^A-Za-z0-9]+/, '')
|
27
|
+
hsh[key].gsub!(/[^A-Za-z0-9]+$/, '')
|
28
|
+
end
|
29
|
+
|
30
|
+
# strip leading numerals
|
31
|
+
# if the real title is quoted inside this string, try to extract it
|
32
|
+
# if the title has at least 2 words before a newline or period or open parens, strip everything after
|
33
|
+
# TODO could do better with knowledge of prepositions, names - maybe we just need a second model?
|
34
|
+
def normalize_title(hsh)
|
35
|
+
str = hsh['title'].strip
|
36
|
+
|
37
|
+
numeral_regexes = [
|
38
|
+
/^[0-9]+[.)](\s+|(?=["'”’´‘“`'A-Z]))/i, # initial numbers + punctuation + space or a quote or a capital letter
|
39
|
+
/^C{0,3}(L?X{0,3}|X[LC])(V?I{0,3}|I[VX])[.)](\s+|(?=["'”’´‘“`'A-Z]))/i, # initial roman numerals
|
40
|
+
/^[A-Z][.)](\s+|(?=["'”’´‘“`'A-Z]))/i # initial single letter
|
41
|
+
]
|
42
|
+
|
43
|
+
numeral_regexes.each do |regex|
|
44
|
+
if str.gsub!(regex, '')
|
45
|
+
break
|
46
|
+
end
|
47
|
+
end
|
48
|
+
|
49
|
+
if (m = str.match /^(["'”’´‘“`'])/)
|
50
|
+
quote_char = m[1]
|
51
|
+
pairable = pairable_quote_chars(quote_char)
|
52
|
+
|
53
|
+
if str.scan(/[#{pairable}]/).length >= 2
|
54
|
+
str.gsub!(/^#{quote_char}/, '')
|
55
|
+
str.gsub!(/[#{pairable}][^#{pairable}]+$/, '')
|
56
|
+
end
|
57
|
+
end
|
58
|
+
|
59
|
+
while (m = str.match /\S+\s+\S+.*(\n|\.(?=\s|\()).*/)
|
60
|
+
i = str.rindex m[1]
|
61
|
+
str = str[0..i-1]
|
62
|
+
end
|
63
|
+
|
64
|
+
hsh['title'] = str
|
65
|
+
normalize('title',hsh)
|
66
|
+
end
|
67
|
+
|
68
|
+
def pairable_quote_chars(quote_char)
|
69
|
+
[%{"”“}, %{’'`‘´'}].each do |chars|
|
70
|
+
return chars if chars.include? quote_char
|
71
|
+
end
|
72
|
+
end
|
73
|
+
|
74
|
+
##
|
75
|
+
# Tries to split the author tokens into individual author names
|
76
|
+
# and then normalizes these names individually. Returns a
|
77
|
+
# list of author names.
|
78
|
+
##
|
79
|
+
def normalize_author(hsh)
|
80
|
+
str = hsh['author']
|
81
|
+
tokens = repair_and_tokenize_author_text(str)
|
82
|
+
authors = []
|
83
|
+
current_auth = []
|
84
|
+
begin_auth = 1
|
85
|
+
tokens.each {|tok|
|
86
|
+
if tok =~ /^(&|and)$/i
|
87
|
+
if !current_auth.empty?
|
88
|
+
auth = normalize_author_name(current_auth)
|
89
|
+
authors << auth
|
90
|
+
end
|
91
|
+
current_auth = []
|
92
|
+
begin_auth = 1
|
93
|
+
next
|
94
|
+
end
|
95
|
+
if begin_auth > 0
|
96
|
+
current_auth << tok
|
97
|
+
begin_auth = 0
|
98
|
+
next
|
99
|
+
end
|
100
|
+
if tok =~ /,$/
|
101
|
+
current_auth << tok
|
102
|
+
if !current_auth.empty?
|
103
|
+
auth = normalize_author_name(current_auth)
|
104
|
+
authors << auth
|
105
|
+
current_auth = []
|
106
|
+
begin_auth = 1
|
107
|
+
end
|
108
|
+
else
|
109
|
+
current_auth << tok
|
110
|
+
end
|
111
|
+
}
|
112
|
+
if !current_auth.empty?
|
113
|
+
auth = normalize_author_name(current_auth)
|
114
|
+
authors << auth.strip unless auth.strip == "-" || auth.strip.blank?
|
115
|
+
end
|
116
|
+
hsh['authors'] = authors if !authors.empty?
|
117
|
+
normalize('author',hsh)
|
118
|
+
hsh
|
119
|
+
end
|
120
|
+
|
121
|
+
def normalize_date(hsh)
|
122
|
+
str = hsh['date']
|
123
|
+
if str =~ /(\d{4})/
|
124
|
+
year = $1.to_i
|
125
|
+
current_year = Time.now.year
|
126
|
+
if year <= current_year+3
|
127
|
+
ret = year
|
128
|
+
hsh['year'] = ret
|
129
|
+
else
|
130
|
+
ret = nil
|
131
|
+
end
|
132
|
+
end
|
133
|
+
hsh['date'] = ret
|
134
|
+
hsh
|
135
|
+
end
|
136
|
+
|
137
|
+
def normalize_volume(hsh)
|
138
|
+
# If there are two numbers, they are volume and number.
|
139
|
+
# e.g. "23(2)", "Vol. 23, No. 3" etc...
|
140
|
+
if hsh['volume'] =~ /\D*(\d+)\D+(\d+)/i
|
141
|
+
hsh['volume'] = $1
|
142
|
+
hsh['number'] = $2
|
143
|
+
# Otherwise, just pull out a number and hope that it's the volume
|
144
|
+
elsif hsh['volume'] =~ /(\d+)/
|
145
|
+
hsh['volume'] = $1
|
146
|
+
end
|
147
|
+
hsh
|
148
|
+
end
|
149
|
+
|
150
|
+
##
|
151
|
+
# Normalizes page fields into the form "start--end". If the page
|
152
|
+
# field does not appear to be in a standard form, does nothing.
|
153
|
+
##
|
154
|
+
def normalize_pages(hsh)
|
155
|
+
# "vol.issue (year):pp"
|
156
|
+
case hsh['pages']
|
157
|
+
when /(\d+) (?: \.(\d+))? (?: \( (\d\d\d\d) \))? : (\d.*)/x
|
158
|
+
hsh['volume'] = $1
|
159
|
+
hsh['number'] = $2 if $2
|
160
|
+
hsh['year'] = $3 if $3
|
161
|
+
hsh['pages'] = $4
|
162
|
+
end
|
163
|
+
|
164
|
+
case hsh['pages']
|
165
|
+
when /(\d+)[^\d]+(\d+)/
|
166
|
+
hsh['pages'] = "#{$1}--#{$2}"
|
167
|
+
when /(\d+)/
|
168
|
+
hsh['pages'] = $1
|
169
|
+
end
|
170
|
+
hsh
|
171
|
+
end
|
172
|
+
|
173
|
+
def repair_and_tokenize_author_text(author_text)
|
174
|
+
# Repair obvious parse errors and weird notations.
|
175
|
+
author_text.sub!(/et\.? al\.?.*$/, '')
|
176
|
+
# FIXME: maybe I'm mis-understanding Perl regular expressions, but
|
177
|
+
# this pattern from ParseCit appears to do the Wrong Thing:
|
178
|
+
# author_text.sub!(/^.*?[a-zA-Z][a-zA-Z]+\. /, '')
|
179
|
+
author_text.gsub!(/\(.*?\)/, '')
|
180
|
+
author_text.gsub!(/^.*?\)\.?/, '')
|
181
|
+
author_text.gsub!(/\(.*?$/, '')
|
182
|
+
author_text.gsub!(/\[.*?\]/, '')
|
183
|
+
author_text.gsub!(/^.*?\]\.?/, '')
|
184
|
+
author_text.gsub!(/\[.*?$/, '')
|
185
|
+
author_text.gsub!(/;/, ',')
|
186
|
+
author_text.gsub!(/,/, ', ')
|
187
|
+
author_text.gsub!(/\:/, ' ')
|
188
|
+
author_text.gsub!(/[\:\"\<\>\/\?\{\}\[\]\+\=\(\)\*\^\%\$\#\@\!\~\_]/, '')
|
189
|
+
author_text = join_multi_word_names(author_text)
|
190
|
+
|
191
|
+
orig_tokens = author_text.split(/\s+/)
|
192
|
+
tokens = []
|
193
|
+
last = false
|
194
|
+
orig_tokens.each_with_index {|tok, i|
|
195
|
+
if tok !~ /[A-Za-z&]/
|
196
|
+
if i < orig_tokens.length/2
|
197
|
+
tokens = []
|
198
|
+
next
|
199
|
+
else
|
200
|
+
last = true
|
201
|
+
end
|
202
|
+
end
|
203
|
+
if (tok =~ /^(jr|sr|ph\.?d|m\.?d|esq)\.?\,?$/i and
|
204
|
+
tokens.last =~ /\,$/) or
|
205
|
+
tok =~ /^[IVX][IVX]+\.?\,?$/
|
206
|
+
|
207
|
+
next
|
208
|
+
end
|
209
|
+
tokens << tok
|
210
|
+
break if last
|
211
|
+
}
|
212
|
+
tokens
|
213
|
+
end # repair_and_tokenize_author_text
|
214
|
+
|
215
|
+
# Insert underscores to join name particles. i.e.
|
216
|
+
# Jon de Groote ---> Jon de_Groote
|
217
|
+
def join_multi_word_names(author_text)
|
218
|
+
author_text.gsub(/\b((?:van|von|der|den|de|di|le|el))\s/i) {
|
219
|
+
"#{$1}_"
|
220
|
+
}
|
221
|
+
end
|
222
|
+
|
223
|
+
##
|
224
|
+
# Tries to normalize an individual author name into the form
|
225
|
+
# "First Middle Last", without punctuation.
|
226
|
+
##
|
227
|
+
def normalize_author_name(auth_toks)
|
228
|
+
return '' if auth_toks.empty?
|
229
|
+
str = auth_toks.join(" ")
|
230
|
+
if str =~ /(.+),\s*(.+)/
|
231
|
+
str = "#{$1} #{$2}"
|
232
|
+
end
|
233
|
+
str.gsub!(/\.\-/, '-')
|
234
|
+
str.gsub!(/[\,\.]/, ' ')
|
235
|
+
str.gsub!(/ +/, ' ')
|
236
|
+
str.strip!
|
237
|
+
|
238
|
+
if (str =~ /^[^\s][^\s]+(\s+[^\s]|\s+[^\s]\-[^\s])+$/)
|
239
|
+
new_toks = str.split(/\s+/)
|
240
|
+
new_order = new_toks[1...new_toks.length];
|
241
|
+
new_order << new_toks[0]
|
242
|
+
str = new_order.join(" ")
|
243
|
+
end
|
244
|
+
|
245
|
+
str.gsub!(/^[^A-Za-z0-9]+/, '')
|
246
|
+
str.gsub!(/[^A-Za-z0-9]+$/, '')
|
247
|
+
return str
|
248
|
+
end
|
249
|
+
|
250
|
+
end
|
251
|
+
|
252
|
+
end
|
@@ -0,0 +1,107 @@
|
|
1
|
+
# encoding: UTF-8
|
2
|
+
|
3
|
+
module Excite
|
4
|
+
|
5
|
+
module Preprocessor
|
6
|
+
|
7
|
+
MARKER_TYPES = {
|
8
|
+
:SQUARE => '\\[.+?\\]',
|
9
|
+
:PAREN => '\\(.+?\\)',
|
10
|
+
:NAKEDNUM => '\\d+',
|
11
|
+
:NAKEDNUMDOT => '\\d+\\.',
|
12
|
+
}
|
13
|
+
|
14
|
+
CLEANUP_RULES_FILE = "#{File.dirname(__FILE__)}/../../config/citation_cleanup_rules.yml"
|
15
|
+
|
16
|
+
def cleanup_rules
|
17
|
+
return @rules if @rules
|
18
|
+
|
19
|
+
raw = YAML.load_file CLEANUP_RULES_FILE
|
20
|
+
@rules = raw['order'].map do |rule_name|
|
21
|
+
re = Regexp.new(raw['rules'][rule_name]['regex'], raw['rules'][rule_name]['ignore_case'])
|
22
|
+
repl = raw['rules'][rule_name]['replacement_str'] || ''
|
23
|
+
{ re: re, repl: repl }
|
24
|
+
end
|
25
|
+
end
|
26
|
+
|
27
|
+
##
|
28
|
+
# Removes lines that appear to be junk from the citation text,
|
29
|
+
# and applies cleanup regexes from the configuration file.
|
30
|
+
##
|
31
|
+
def normalize_cite_text(cite_text)
|
32
|
+
cite_text.split(/\n/).reject do |line|
|
33
|
+
line.blank? || line =~ /^[\s\d]*$/
|
34
|
+
end.map do |line|
|
35
|
+
normalize_citation(line)
|
36
|
+
end.join("\n")
|
37
|
+
end
|
38
|
+
|
39
|
+
def normalize_citation(cite)
|
40
|
+
cite = cite.dup
|
41
|
+
|
42
|
+
cleanup_rules.each do |rule|
|
43
|
+
cite.gsub!(rule[:re], rule[:repl])
|
44
|
+
end
|
45
|
+
|
46
|
+
cite
|
47
|
+
end
|
48
|
+
|
49
|
+
##
|
50
|
+
# Controls the process by which citations are segmented,
|
51
|
+
# based on the result of trying to guess the type of
|
52
|
+
# citation marker used in the reference section. Returns
|
53
|
+
# a reference to a list of citation objects.
|
54
|
+
##
|
55
|
+
def segment_citations(cite_text)
|
56
|
+
marker_type = guess_marker_type(cite_text)
|
57
|
+
unless marker_type == 'UNKNOWN'
|
58
|
+
citations = split_unmarked_citations(cite_text)
|
59
|
+
else
|
60
|
+
citations = split_citations_by_marker(cite_text, marker_type)
|
61
|
+
end
|
62
|
+
return citations
|
63
|
+
end
|
64
|
+
|
65
|
+
##
|
66
|
+
# Segments citations that have explicit markers in the
|
67
|
+
# reference section. Whenever a new line starts with an
|
68
|
+
# expression that matches what we'd expect of a marker,
|
69
|
+
# a new citation is started. Returns a reference to a
|
70
|
+
# list of citation objects.
|
71
|
+
##
|
72
|
+
def split_citations_by_marker(cite_text, marker_type=nil)
|
73
|
+
citations = []
|
74
|
+
current_citation = Citation.new
|
75
|
+
current_citation_string = nil
|
76
|
+
|
77
|
+
cite_text.split(/\n/).each {|line|
|
78
|
+
if line =~ /^\s*(#{MARKER_TYPES{marker_type}})\s*(.*)$/
|
79
|
+
marker, cite_string = $1, $2
|
80
|
+
if current_citation_string
|
81
|
+
current_citation.citation_string = current_citation_string
|
82
|
+
citations << current_citation
|
83
|
+
current_citation_string = nil
|
84
|
+
end
|
85
|
+
current_citation = Citation.new
|
86
|
+
current_citation.marker_type = marker_type
|
87
|
+
current_citation.marker = marker
|
88
|
+
current_citation_string = cite_string
|
89
|
+
else
|
90
|
+
if current_citation_string =~ /\s\-$/
|
91
|
+
current_citation_string.sub(/\-$/, '')
|
92
|
+
current_citation_string << line
|
93
|
+
else
|
94
|
+
current_citation_string << " " << line
|
95
|
+
end
|
96
|
+
end
|
97
|
+
}
|
98
|
+
|
99
|
+
if current_citation && current_citation_string
|
100
|
+
current_citation.string = current_citation_string
|
101
|
+
citations << current_citation
|
102
|
+
end
|
103
|
+
citations
|
104
|
+
end
|
105
|
+
|
106
|
+
end
|
107
|
+
end
|