gabrielaltay
commited on
Commit
•
90131cc
1
Parent(s):
e5d0ba0
upload hubscripts/nlm_wsd_hub.py to hub from bigbio repo
Browse files- nlm_wsd.py +362 -0
nlm_wsd.py
ADDED
@@ -0,0 +1,362 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# coding=utf-8
|
2 |
+
# Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
|
3 |
+
#
|
4 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
+
# you may not use this file except in compliance with the License.
|
6 |
+
# You may obtain a copy of the License at
|
7 |
+
#
|
8 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
9 |
+
#
|
10 |
+
# Unless required by applicable law or agreed to in writing, software
|
11 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
12 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
13 |
+
# See the License for the specific language governing permissions and
|
14 |
+
# limitations under the License.
|
15 |
+
|
16 |
+
"""
|
17 |
+
In order to support research investigating the automatic resolution of word sense ambiguity using natural language
|
18 |
+
processing techniques, we have constructed this test collection of medical text in which the ambiguities were resolved
|
19 |
+
by hand. Evaluators were asked to examine instances of an ambiguous word and determine the sense intended by selecting
|
20 |
+
the Metathesaurus concept (if any) that best represents the meaning of that sense. The test collection consists of 50
|
21 |
+
highly frequent ambiguous UMLS concepts from 1998 MEDLINE. Each of the 50 ambiguous cases has 100 ambiguous instances
|
22 |
+
randomly selected from the 1998 MEDLINE citations. For a total of 5,000 instances. We had a total of 11 evaluators of
|
23 |
+
which 8 completed 100% of the 5,000 instances, 1 completed 56%, 1 completed 44%, and the final evaluator completed 12%
|
24 |
+
of the instances. Evaluations were only used when the evaluators completed all 100 instances for a given ambiguity.
|
25 |
+
|
26 |
+
Comment from author:
|
27 |
+
BigBio schema fixes off by one error of end offset of entities. The source config remains unchanged.
|
28 |
+
|
29 |
+
Instructions on how to load locally:
|
30 |
+
1) Create directory
|
31 |
+
2) Download one of the following annotation sets from https://lhncbc.nlm.nih.gov/restricted/ii/areas/WSD/index.html
|
32 |
+
and put it into the folder:
|
33 |
+
- Full Reviewed Set
|
34 |
+
https://lhncbc.nlm.nih.gov/restricted/ii/areas/WSD/downloads/full_reviewed_results.tar.gz
|
35 |
+
(Link "Full Reviewed Result Set (requires Common Files above)")
|
36 |
+
subset_id = nlm_wsd_reviewed
|
37 |
+
- Full Non-Reviewed Set
|
38 |
+
https://lhncbc.nlm.nih.gov/restricted/ii/areas/WSD/downloads/full_non_reviewed_results.tar.gz
|
39 |
+
(Link "Full Non-Reviewed Result Set (requires Common Files above)")
|
40 |
+
subset_id = nlm_wsd_non_reviewed
|
41 |
+
3) Download https://lhncbc.nlm.nih.gov/restricted/ii/areas/WSD/downloads/UMLS1999.tar.gz (Link "1999 UMLS Data Files")
|
42 |
+
and put it into the folder
|
43 |
+
4) Set kwarg data_dir of load_datasets to the path of the directory
|
44 |
+
"""
|
45 |
+
|
46 |
+
import itertools as it
|
47 |
+
import re
|
48 |
+
from dataclasses import dataclass
|
49 |
+
from pathlib import Path
|
50 |
+
from typing import Dict, List, Tuple
|
51 |
+
|
52 |
+
import datasets
|
53 |
+
|
54 |
+
from .bigbiohub import kb_features
|
55 |
+
from .bigbiohub import BigBioConfig
|
56 |
+
from .bigbiohub import Tasks
|
57 |
+
|
58 |
+
_LANGUAGES = ['English']
|
59 |
+
_PUBMED = True
|
60 |
+
_LOCAL = True
|
61 |
+
_CITATION = """\
|
62 |
+
@article{weeber2001developing,
|
63 |
+
title = "Developing a test collection for biomedical word sense
|
64 |
+
disambiguation",
|
65 |
+
author = "Weeber, M and Mork, J G and Aronson, A R",
|
66 |
+
journal = "Proc AMIA Symp",
|
67 |
+
pages = "746--750",
|
68 |
+
year = 2001,
|
69 |
+
language = "en"
|
70 |
+
}
|
71 |
+
"""
|
72 |
+
|
73 |
+
_DATASETNAME = "nlm_wsd"
|
74 |
+
_DISPLAYNAME = "NLM WSD"
|
75 |
+
|
76 |
+
_DESCRIPTION = """\
|
77 |
+
In order to support research investigating the automatic resolution of word sense ambiguity using natural language
|
78 |
+
processing techniques, we have constructed this test collection of medical text in which the ambiguities were resolved
|
79 |
+
by hand. Evaluators were asked to examine instances of an ambiguous word and determine the sense intended by selecting
|
80 |
+
the Metathesaurus concept (if any) that best represents the meaning of that sense. The test collection consists of 50
|
81 |
+
highly frequent ambiguous UMLS concepts from 1998 MEDLINE. Each of the 50 ambiguous cases has 100 ambiguous instances
|
82 |
+
randomly selected from the 1998 MEDLINE citations. For a total of 5,000 instances. We had a total of 11 evaluators of
|
83 |
+
which 8 completed 100% of the 5,000 instances, 1 completed 56%, 1 completed 44%, and the final evaluator completed 12%
|
84 |
+
of the instances. Evaluations were only used when the evaluators completed all 100 instances for a given ambiguity.
|
85 |
+
"""
|
86 |
+
|
87 |
+
_HOMEPAGE = "https://lhncbc.nlm.nih.gov/restricted/ii/areas/WSD/index.html"
|
88 |
+
|
89 |
+
_LICENSE = 'UMLS - Metathesaurus License Agreement'
|
90 |
+
|
91 |
+
_URLS = {
|
92 |
+
"UMLS": "UMLS1999.tar.gz",
|
93 |
+
"reviewed": "full_reviewed_results.tar.gz",
|
94 |
+
"non_reviewed": "full_non_reviewed_results.tar.gz",
|
95 |
+
}
|
96 |
+
|
97 |
+
_SUPPORTED_TASKS = [Tasks.NAMED_ENTITY_DISAMBIGUATION]
|
98 |
+
|
99 |
+
_SOURCE_VERSION = "1.0.0"
|
100 |
+
_BIGBIO_VERSION = "1.0.0"
|
101 |
+
|
102 |
+
|
103 |
+
@dataclass
|
104 |
+
class NlmWsdBigBioConfig(BigBioConfig):
|
105 |
+
schema: str = "source"
|
106 |
+
name: str = "nlm_wsd_reviewed_source"
|
107 |
+
version: datasets.Version = datasets.Version(_SOURCE_VERSION)
|
108 |
+
description: str = "NLM-WSD basic reviewed source schema"
|
109 |
+
subset_id: str = "nlm_wsd_reviewed"
|
110 |
+
|
111 |
+
|
112 |
+
class NlmWsdDataset(datasets.GeneratorBasedBuilder):
|
113 |
+
"""Biomedical Word Sense Disambiguation (WSD)."""
|
114 |
+
|
115 |
+
uid = it.count(0)
|
116 |
+
|
117 |
+
SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
|
118 |
+
BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION)
|
119 |
+
|
120 |
+
BUILDER_CONFIGS = [
|
121 |
+
NlmWsdBigBioConfig(
|
122 |
+
name="nlm_wsd_non_reviewed_source",
|
123 |
+
version=SOURCE_VERSION,
|
124 |
+
description="NLM-WSD basic non reviewed source schema",
|
125 |
+
schema="source",
|
126 |
+
subset_id="nlm_wsd_non_reviewed",
|
127 |
+
),
|
128 |
+
NlmWsdBigBioConfig(
|
129 |
+
name="nlm_wsd_non_reviewed_bigbio_kb",
|
130 |
+
version=BIGBIO_VERSION,
|
131 |
+
description="NLM-WSD basic non reviewed BigBio schema",
|
132 |
+
schema="bigbio_kb",
|
133 |
+
subset_id="nlm_wsd_non_reviewed",
|
134 |
+
),
|
135 |
+
NlmWsdBigBioConfig(
|
136 |
+
name="nlm_wsd_reviewed_source",
|
137 |
+
version=SOURCE_VERSION,
|
138 |
+
description="NLM-WSD basic reviewed source schema",
|
139 |
+
schema="source",
|
140 |
+
subset_id="nlm_wsd_reviewed",
|
141 |
+
),
|
142 |
+
NlmWsdBigBioConfig(
|
143 |
+
name="nlm_wsd_reviewed_bigbio_kb",
|
144 |
+
version=BIGBIO_VERSION,
|
145 |
+
description="NLM-WSD basic reviewed BigBio schema",
|
146 |
+
schema="bigbio_kb",
|
147 |
+
subset_id="nlm_wsd_reviewed",
|
148 |
+
),
|
149 |
+
]
|
150 |
+
|
151 |
+
BUILDER_CONFIG_CLASS = NlmWsdBigBioConfig
|
152 |
+
|
153 |
+
def _info(self) -> datasets.DatasetInfo:
|
154 |
+
if self.config.schema == "source":
|
155 |
+
features = datasets.Features(
|
156 |
+
{
|
157 |
+
"id": datasets.Value("string"),
|
158 |
+
"sentence_id": datasets.Value("string"),
|
159 |
+
"label": datasets.Value("string"),
|
160 |
+
"sentence": {
|
161 |
+
"text": datasets.Value("string"),
|
162 |
+
"ambiguous_word": datasets.Value("string"),
|
163 |
+
"ambiguous_word_alias": datasets.Value("string"),
|
164 |
+
"offsets_context": datasets.Sequence(datasets.Value("int32")),
|
165 |
+
"offsets_ambiguity": datasets.Sequence(datasets.Value("int32")),
|
166 |
+
"context": datasets.Value("string"),
|
167 |
+
},
|
168 |
+
"citation": {
|
169 |
+
"text": datasets.Value("string"),
|
170 |
+
"ambiguous_word": datasets.Value("string"),
|
171 |
+
"ambiguous_word_alias": datasets.Value("string"),
|
172 |
+
"offsets_context": datasets.Sequence(datasets.Value("int32")),
|
173 |
+
"offsets_ambiguity": datasets.Sequence(datasets.Value("int32")),
|
174 |
+
"context": datasets.Value("string"),
|
175 |
+
},
|
176 |
+
"choices": [
|
177 |
+
{
|
178 |
+
"label": datasets.Value("string"),
|
179 |
+
"concept": datasets.Value("string"),
|
180 |
+
"cui": datasets.Value("string"),
|
181 |
+
"type": [datasets.Value("string")],
|
182 |
+
}
|
183 |
+
],
|
184 |
+
}
|
185 |
+
)
|
186 |
+
elif self.config.schema == "bigbio_kb":
|
187 |
+
features = kb_features
|
188 |
+
|
189 |
+
return datasets.DatasetInfo(
|
190 |
+
description=_DESCRIPTION,
|
191 |
+
features=features,
|
192 |
+
homepage=_HOMEPAGE,
|
193 |
+
license=str(_LICENSE),
|
194 |
+
citation=_CITATION,
|
195 |
+
)
|
196 |
+
|
197 |
+
def _split_generators(self, dl_manager) -> List[datasets.SplitGenerator]:
|
198 |
+
"""Returns SplitGenerators."""
|
199 |
+
|
200 |
+
if self.config.data_dir is None:
|
201 |
+
raise ValueError(
|
202 |
+
"This is a local dataset. Please pass the data_dir kwarg to load_dataset."
|
203 |
+
)
|
204 |
+
else:
|
205 |
+
data_dir = Path(self.config.data_dir)
|
206 |
+
umls_dir = dl_manager.download_and_extract(data_dir / _URLS["UMLS"])
|
207 |
+
mrcon_path = Path(umls_dir) / "META" / "MRCON"
|
208 |
+
if self.config.subset_id == "nlm_wsd_reviewed":
|
209 |
+
ann_dir = dl_manager.download_and_extract(data_dir / _URLS["reviewed"])
|
210 |
+
ann_dir = Path(ann_dir) / "Reviewed_Results"
|
211 |
+
else:
|
212 |
+
ann_dir = dl_manager.download_and_extract(
|
213 |
+
data_dir / _URLS["non_reviewed"]
|
214 |
+
)
|
215 |
+
ann_dir = Path(ann_dir) / "Non-Reviewed_Results"
|
216 |
+
|
217 |
+
return [
|
218 |
+
datasets.SplitGenerator(
|
219 |
+
name=datasets.Split.TRAIN,
|
220 |
+
gen_kwargs={
|
221 |
+
"mrcon_path": mrcon_path,
|
222 |
+
"ann_dir": ann_dir,
|
223 |
+
},
|
224 |
+
)
|
225 |
+
]
|
226 |
+
|
227 |
+
def _generate_examples(self, mrcon_path: Path, ann_dir: Path) -> Tuple[int, Dict]:
|
228 |
+
"""Yields examples as (key, example) tuples."""
|
229 |
+
|
230 |
+
# read label->cui map
|
231 |
+
umls_map = {}
|
232 |
+
with mrcon_path.open() as f:
|
233 |
+
content = f.readlines()
|
234 |
+
content = [x.strip() for x in content]
|
235 |
+
for line in content:
|
236 |
+
fields = line.split("|")
|
237 |
+
assert len(fields) == 9, f"{len(fields)}"
|
238 |
+
assert fields[0][0] == "C"
|
239 |
+
umls_map[fields[6]] = fields[0]
|
240 |
+
|
241 |
+
for dir in ann_dir.iterdir():
|
242 |
+
if self.config.schema == "source" and dir.is_dir():
|
243 |
+
for example in self._generate_parsed_documents(dir, umls_map):
|
244 |
+
yield next(self.uid), example
|
245 |
+
|
246 |
+
elif self.config.schema == "bigbio_kb" and dir.is_dir():
|
247 |
+
for example in self._generate_parsed_documents(dir, umls_map):
|
248 |
+
yield next(self.uid), self._source_to_kb(example)
|
249 |
+
|
250 |
+
def _generate_parsed_documents(self, dir, umls_map):
|
251 |
+
|
252 |
+
# read choices
|
253 |
+
choices = []
|
254 |
+
choices_path = dir / "choices"
|
255 |
+
with choices_path.open() as f:
|
256 |
+
content = f.readlines()
|
257 |
+
content = [x.strip() for x in content]
|
258 |
+
for line in content:
|
259 |
+
label, concept, *type = line.split("|")
|
260 |
+
type = [x.split(", ")[1] for x in type]
|
261 |
+
m = re.search(r"(?<=\().+(?=\))", concept)
|
262 |
+
if m is None:
|
263 |
+
choices.append(
|
264 |
+
{"label": label, "concept": concept, "type": type, "cui": ""}
|
265 |
+
)
|
266 |
+
else:
|
267 |
+
concept = m.group()
|
268 |
+
choices.append(
|
269 |
+
{
|
270 |
+
"label": label,
|
271 |
+
"concept": concept,
|
272 |
+
"type": type,
|
273 |
+
"cui": umls_map[concept],
|
274 |
+
}
|
275 |
+
)
|
276 |
+
|
277 |
+
file_path = dir / f"{dir.name}_set"
|
278 |
+
with file_path.open() as f:
|
279 |
+
for raw_document in self._generate_raw_documents(f):
|
280 |
+
document = {}
|
281 |
+
id, document_id, label = raw_document[0].strip().split("|")
|
282 |
+
|
283 |
+
info_sentence = self._parse_ambig_pos_info(raw_document[2].strip())
|
284 |
+
info_sentence["text"] = raw_document[1]
|
285 |
+
|
286 |
+
info_citation = self._parse_ambig_pos_info(raw_document[-1].strip())
|
287 |
+
n_cit = len(raw_document) - 3
|
288 |
+
info_citation["text"] = "".join(raw_document[3 : 3 + n_cit])
|
289 |
+
|
290 |
+
document = {
|
291 |
+
"id": id,
|
292 |
+
"sentence_id": document_id,
|
293 |
+
"label": label,
|
294 |
+
"sentence": info_sentence,
|
295 |
+
"citation": info_citation,
|
296 |
+
"choices": choices,
|
297 |
+
}
|
298 |
+
yield document
|
299 |
+
|
300 |
+
def _generate_raw_documents(self, fstream):
|
301 |
+
raw_document = []
|
302 |
+
for line in fstream:
|
303 |
+
if line.strip():
|
304 |
+
raw_document.append(line)
|
305 |
+
elif raw_document:
|
306 |
+
yield raw_document
|
307 |
+
raw_document = []
|
308 |
+
# needed for last document
|
309 |
+
if raw_document:
|
310 |
+
yield raw_document
|
311 |
+
|
312 |
+
def _parse_ambig_pos_info(self, line):
|
313 |
+
infos = line.split("|")
|
314 |
+
assert len(infos) == 8, f"{len(infos)}"
|
315 |
+
pos_info = {
|
316 |
+
"ambiguous_word": infos[0],
|
317 |
+
"ambiguous_word_alias": infos[1],
|
318 |
+
"offsets_context": [infos[2], infos[3]],
|
319 |
+
"offsets_ambiguity": [infos[4], infos[5]],
|
320 |
+
"context": infos[6],
|
321 |
+
}
|
322 |
+
return pos_info
|
323 |
+
|
324 |
+
def _source_to_kb(self, example):
|
325 |
+
document_ = {}
|
326 |
+
document_["events"] = []
|
327 |
+
document_["relations"] = []
|
328 |
+
document_["coreferences"] = []
|
329 |
+
document_["id"] = next(self.uid)
|
330 |
+
document_["document_id"] = example["sentence_id"].split(".")[0]
|
331 |
+
|
332 |
+
citation = example["citation"]
|
333 |
+
document_["passages"] = [
|
334 |
+
{
|
335 |
+
"id": next(self.uid),
|
336 |
+
"type": "",
|
337 |
+
"text": [citation["text"]],
|
338 |
+
"offsets": [[0, len(citation["text"])]],
|
339 |
+
}
|
340 |
+
]
|
341 |
+
choices = {x["label"]: x["cui"] for x in example["choices"]}
|
342 |
+
types = {x["label"]: x["type"][0] for x in example["choices"]}
|
343 |
+
|
344 |
+
db_id = (
|
345 |
+
"" if example["label"] in ["None", "UNDEF"] else choices[example["label"]]
|
346 |
+
)
|
347 |
+
type = "" if example["label"] in ["None", "UNDEF"] else types[example["label"]]
|
348 |
+
document_["entities"] = [
|
349 |
+
{
|
350 |
+
"id": next(self.uid),
|
351 |
+
"type": type,
|
352 |
+
"text": [citation["ambiguous_word_alias"]],
|
353 |
+
"offsets": [
|
354 |
+
[
|
355 |
+
int(citation["offsets_ambiguity"][0]),
|
356 |
+
int(citation["offsets_ambiguity"][1]) + 1,
|
357 |
+
]
|
358 |
+
],
|
359 |
+
"normalized": [{"db_name": "UMLS", "db_id": db_id}],
|
360 |
+
}
|
361 |
+
]
|
362 |
+
return document_
|