/
negation.py
237 lines (205 loc) · 8.33 KB
/
negation.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
from spacy.tokens import Token, Doc, Span
from spacy.matcher import PhraseMatcher
import logging
from negspacy.termsets import LANGUAGES
class Negex:
"""
A spaCy pipeline component which identifies negated tokens in text.
Based on: NegEx - A Simple Algorithm for Identifying Negated Findings and Diseasesin Discharge Summaries
Chapman, Bridewell, Hanbury, Cooper, Buchanan
Parameters
----------
nlp: object
spaCy language object
ent_types: list
list of entity types to negate
language: str
language code, if using default termsets (e.g. "en" for english)
psuedo_negations: list
list of phrases that cancel out a negation, if empty, defaults are used
preceding_negations: list
negations that appear before an entity, if empty, defaults are used
following_negations: list
negations that appear after an entity, if empty, defaults are used
termination: list
phrases that "terminate" a sentence for processing purposes such as "but". If empty, defaults are used
"""
def __init__(
self,
nlp,
language="en",
ent_types=list(),
psuedo_negations=list(),
preceding_negations=list(),
following_negations=list(),
termination=list(),
chunk_prefix=list(),
):
if not language in LANGUAGES:
raise KeyError(
f"{language} not found in languages termset. "
"Ensure this is a supported language or specify "
"your own termsets when initializing Negex."
)
termsets = LANGUAGES[language]
if not Span.has_extension("negex"):
Span.set_extension("negex", default=False, force=True)
if not psuedo_negations:
if not "psuedo_negations" in termsets:
raise KeyError("psuedo_negations not specified for this language.")
psuedo_negations = termsets["psuedo_negations"]
if not preceding_negations:
if not "preceding_negations" in termsets:
raise KeyError("preceding_negations not specified for this language.")
preceding_negations = termsets["preceding_negations"]
if not following_negations:
if not "following_negations" in termsets:
raise KeyError("following_negations not specified for this language.")
following_negations = termsets["following_negations"]
if not termination:
if not "termination" in termsets:
raise KeyError("termination not specified for this language.")
termination = termsets["termination"]
# efficiently build spaCy matcher patterns
self.psuedo_patterns = list(nlp.tokenizer.pipe(psuedo_negations))
self.preceding_patterns = list(nlp.tokenizer.pipe(preceding_negations))
self.following_patterns = list(nlp.tokenizer.pipe(following_negations))
self.termination_patterns = list(nlp.tokenizer.pipe(termination))
self.matcher = PhraseMatcher(nlp.vocab, attr="LOWER")
self.matcher.add("Psuedo", None, *self.psuedo_patterns)
self.matcher.add("Preceding", None, *self.preceding_patterns)
self.matcher.add("Following", None, *self.following_patterns)
self.matcher.add("Termination", None, *self.termination_patterns)
self.nlp = nlp
self.ent_types = ent_types
self.chunk_prefix = list(nlp.tokenizer.pipe(chunk_prefix))
def get_patterns(self):
"""
returns phrase patterns used for various negation dictionaries
Returns
-------
patterns: dict
pattern_type: [patterns]
"""
patterns = {
"psuedo_patterns": self.psuedo_patterns,
"preceding_patterns": self.preceding_patterns,
"following_patterns": self.following_patterns,
"termination_patterns": self.termination_patterns,
}
for pattern in patterns:
logging.info(pattern)
return patterns
def process_negations(self, doc):
"""
Find negations in doc and clean candidate negations to remove pseudo negations
Parameters
----------
doc: object
spaCy Doc object
Returns
-------
preceding: list
list of tuples for preceding negations
following: list
list of tuples for following negations
terminating: list
list of tuples of terminating phrases
"""
###
# does not work properly in spacy 2.1.8. Will incorporate after 2.2.
# Relying on user to use NER in meantime
# see https://github.com/jenojp/negspacy/issues/7
###
# if not doc.is_nered:
# raise ValueError(
# "Negations are evaluated for Named Entities found in text. "
# "Your SpaCy pipeline does not included Named Entity resolution. "
# "Please ensure it is enabled or choose a different language model that includes it."
# )
preceding = list()
following = list()
terminating = list()
matches = self.matcher(doc)
psuedo = [
(match_id, start, end)
for match_id, start, end in matches
if self.nlp.vocab.strings[match_id] == "Psuedo"
]
for match_id, start, end in matches:
if self.nlp.vocab.strings[match_id] == "Psuedo":
continue
psuedo_flag = False
for p in psuedo:
if start >= p[1] and start <= p[2]:
psuedo_flag == True
continue
if not psuedo_flag:
if self.nlp.vocab.strings[match_id] == "Preceding":
preceding.append((match_id, start, end))
elif self.nlp.vocab.strings[match_id] == "Following":
following.append((match_id, start, end))
elif self.nlp.vocab.strings[match_id] == "Termination":
terminating.append((match_id, start, end))
else:
logging.warnings(
f"phrase {doc[start:end].text} not in one of the expected matcher types."
)
return preceding, following, terminating
def termination_boundaries(self, doc, terminating):
"""
Create sub sentences based on terminations found in text.
Parameters
----------
doc: object
spaCy Doc object
terminating: list
list of tuples with (match_id, start, end)
returns
-------
boundaries: list
list of tuples with (start, end) of spans
"""
sent_starts = [sent.start for sent in doc.sents]
terminating_starts = [t[1] for t in terminating]
starts = sent_starts + terminating_starts + [len(doc)]
starts.sort()
boundaries = list()
index = 0
for i, start in enumerate(starts):
if not i == 0:
boundaries.append((index, start))
index = start
return boundaries
def negex(self, doc):
"""
Negates entities of interest
Parameters
----------
doc: object
spaCy Doc object
"""
preceding, following, terminating = self.process_negations(doc)
boundaries = self.termination_boundaries(doc, terminating)
for b in boundaries:
sub_preceding = [i for i in preceding if b[0] <= i[1] < b[1]]
sub_following = [i for i in following if b[0] <= i[1] < b[1]]
for e in doc[b[0] : b[1]].ents:
if self.ent_types:
if e.label_ not in self.ent_types:
continue
if any(pre < e.start for pre in [i[1] for i in sub_preceding]):
e._.negex = True
continue
if any(fol > e.end for fol in [i[2] for i in sub_following]):
e._.negex = True
continue
if self.chunk_prefix:
if any(
c.text.lower() == doc[e.start].text.lower()
for c in self.chunk_prefix
):
e._.negex = True
return doc
def __call__(self, doc):
return self.negex(doc)