qanastek commited on
Commit
b829cac
1 Parent(s): d7fa5f6

Upload 2 files

Browse files
Files changed (2) hide show
  1. E3C.py +229 -0
  2. test_e3c.py +8 -0
E3C.py ADDED
@@ -0,0 +1,229 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # pip install bs4 syntok
2
+
3
+ import os
4
+ import random
5
+
6
+ import datasets
7
+
8
+ import numpy as np
9
+ from bs4 import BeautifulSoup, ResultSet
10
+ from syntok.tokenizer import Tokenizer
11
+
12
+ tokenizer = Tokenizer()
13
+
14
+ _CITATION = """\
15
+ @report{Magnini2021, \
16
+ author = {Bernardo Magnini and Begoña Altuna and Alberto Lavelli and Manuela Speranza \
17
+ and Roberto Zanoli and Fondazione Bruno Kessler}, \
18
+ keywords = {Clinical data,clinical enti-ties,corpus,multilingual,temporal information}, \
19
+ title = {The E3C Project: \
20
+ European Clinical Case Corpus El proyecto E3C: European Clinical Case Corpus}, \
21
+ url = {https://uts.nlm.nih.gov/uts/umls/home}, \
22
+ year = {2021}, \
23
+ }
24
+ """
25
+
26
+ _DESCRIPTION = """\
27
+ E3C is a freely available multilingual corpus (English, French, Italian, Spanish, and Basque) \
28
+ of semantically annotated clinical narratives to allow for the linguistic analysis, benchmarking, \
29
+ and training of information extraction systems. It consists of two types of annotations: \
30
+ (i) clinical entities (e.g., pathologies), (ii) temporal information and factuality (e.g., events). \
31
+ Researchers can use the benchmark training and test splits of our corpus to develop and test \
32
+ their own models.
33
+ """
34
+
35
+ _URL = "https://github.com/hltfbk/E3C-Corpus/archive/refs/tags/v2.0.0.zip"
36
+
37
+ _LANGUAGES = ["English","Spanish","Basque","French","Italian"]
38
+
39
+ class E3C(datasets.GeneratorBasedBuilder):
40
+
41
+ BUILDER_CONFIGS = [
42
+ datasets.BuilderConfig(name=f"{lang}", version="1.0.0", description=f"The {lang} subset of the E3C corpus") for lang in _LANGUAGES
43
+ ]
44
+
45
+ def _info(self):
46
+
47
+ features = datasets.Features(
48
+ {
49
+ "id": datasets.Value("string"),
50
+ "text": datasets.Value("string"),
51
+ "tokens": datasets.Sequence(datasets.Value("string")),
52
+ "ner_clinical_tags": datasets.Sequence(
53
+ datasets.features.ClassLabel(
54
+ names=["O","B-CLINENTITY","I-CLINENTITY"],
55
+ ),
56
+ ),
57
+ "ner_temporal_tags": datasets.Sequence(
58
+ datasets.features.ClassLabel(
59
+ names=["O", "B-EVENT", "B-ACTOR", "B-BODYPART", "B-TIMEX3", "B-RML", "I-EVENT", "I-ACTOR", "I-BODYPART", "I-TIMEX3", "I-RML"],
60
+ ),
61
+ ),
62
+ }
63
+ )
64
+
65
+ return datasets.DatasetInfo(
66
+ description=_DESCRIPTION,
67
+ features=features,
68
+ citation=_CITATION,
69
+ supervised_keys=None,
70
+ )
71
+
72
+ def _split_generators(self, dl_manager):
73
+
74
+ data_dir = dl_manager.download_and_extract(_URL)
75
+
76
+ return [
77
+ datasets.SplitGenerator(
78
+ name=datasets.Split.TRAIN,
79
+ gen_kwargs={
80
+ "filepath": os.path.join(data_dir, "E3C-Corpus-2.0.0/data_annotation", self.config.name, "layer1"),
81
+ "split": "train",
82
+ },
83
+ ),
84
+ datasets.SplitGenerator(
85
+ name=datasets.Split.VALIDATION,
86
+ gen_kwargs={
87
+ "filepath": os.path.join(data_dir, "E3C-Corpus-2.0.0/data_annotation", self.config.name, "layer1"),
88
+ "split": "validation",
89
+ },
90
+ ),
91
+ datasets.SplitGenerator(
92
+ name=datasets.Split.TEST,
93
+ gen_kwargs={
94
+ "filepath": os.path.join(data_dir, "E3C-Corpus-2.0.0/data_validation", self.config.name, "layer2"),
95
+ "split": "test",
96
+ },
97
+ ),
98
+ ]
99
+
100
+ @staticmethod
101
+ def get_annotations(entities: ResultSet, text: str) -> list:
102
+
103
+ return [[
104
+ int(entity.get("begin")),
105
+ int(entity.get("end")),
106
+ text[int(entity.get("begin")) : int(entity.get("end"))],
107
+ ] for entity in entities]
108
+
109
+ def get_clinical_annotations(self, entities: ResultSet, text: str) -> list:
110
+
111
+ return [[
112
+ int(entity.get("begin")),
113
+ int(entity.get("end")),
114
+ text[int(entity.get("begin")) : int(entity.get("end"))],
115
+ entity.get("entityID"),
116
+ ] for entity in entities]
117
+
118
+ def get_parsed_data(self, filepath: str):
119
+
120
+ for root, _, files in os.walk(filepath):
121
+
122
+ for file in files:
123
+
124
+ with open(f"{root}/{file}") as soup_file:
125
+
126
+ soup = BeautifulSoup(soup_file, "xml")
127
+ text = soup.find("cas:Sofa").get("sofaString")
128
+
129
+ yield {
130
+ "CLINENTITY": self.get_clinical_annotations(soup.find_all("custom:CLINENTITY"), text),
131
+ "EVENT": self.get_annotations(soup.find_all("custom:EVENT"), text),
132
+ "ACTOR": self.get_annotations(soup.find_all("custom:ACTOR"), text),
133
+ "BODYPART": self.get_annotations(soup.find_all("custom:BODYPART"), text),
134
+ "TIMEX3": self.get_annotations(soup.find_all("custom:TIMEX3"), text),
135
+ "RML": self.get_annotations(soup.find_all("custom:RML"), text),
136
+ "SENTENCE": self.get_annotations(soup.find_all("type4:Sentence"), text),
137
+ "TOKENS": self.get_annotations(soup.find_all("type4:Token"), text),
138
+ }
139
+
140
+ def _generate_examples(self, filepath, split):
141
+
142
+ all_res = []
143
+
144
+ key = 0
145
+
146
+ for content in self.get_parsed_data(filepath):
147
+
148
+ for sentence in content["SENTENCE"]:
149
+
150
+ tokens = [(
151
+ token.offset + sentence[0],
152
+ token.offset + sentence[0] + len(token.value),
153
+ token.value,
154
+ ) for token in list(tokenizer.tokenize(sentence[-1]))]
155
+
156
+ filtered_tokens = list(
157
+ filter(
158
+ lambda token: token[0] >= sentence[0] and token[1] <= sentence[1],
159
+ tokens,
160
+ )
161
+ )
162
+
163
+ tokens_offsets = [
164
+ [token[0] - sentence[0], token[1] - sentence[0]] for token in filtered_tokens
165
+ ]
166
+
167
+ clinical_labels = ["O"] * len(filtered_tokens)
168
+ clinical_cuid = ["CUI_LESS"] * len(filtered_tokens)
169
+ temporal_information_labels = ["O"] * len(filtered_tokens)
170
+
171
+ for entity_type in ["CLINENTITY","EVENT","ACTOR","BODYPART","TIMEX3","RML"]:
172
+
173
+ if len(content[entity_type]) != 0:
174
+
175
+ for entities in list(content[entity_type]):
176
+
177
+ annotated_tokens = [
178
+ idx_token
179
+ for idx_token, token in enumerate(filtered_tokens)
180
+ if token[0] >= entities[0] and token[1] <= entities[1]
181
+ ]
182
+
183
+ for idx_token in annotated_tokens:
184
+
185
+ if entity_type == "CLINENTITY":
186
+ if idx_token == annotated_tokens[0]:
187
+ clinical_labels[idx_token] = f"B-{entity_type}"
188
+ else:
189
+ clinical_labels[idx_token] = f"I-{entity_type}"
190
+ clinical_cuid[idx_token] = entities[-1]
191
+ else:
192
+ if idx_token == annotated_tokens[0]:
193
+ temporal_information_labels[idx_token] = f"B-{entity_type}"
194
+ else:
195
+ temporal_information_labels[idx_token] = f"I-{entity_type}"
196
+
197
+ all_res.append({
198
+ "id": key,
199
+ "text": sentence[-1],
200
+ "tokens": list(map(lambda token: token[2], filtered_tokens)),
201
+ "ner_clinical_tags": clinical_labels,
202
+ "ner_temporal_tags": temporal_information_labels,
203
+ })
204
+
205
+ key += 1
206
+
207
+ if split != "test":
208
+
209
+ ids = [r["id"] for r in all_res]
210
+
211
+ random.seed(4)
212
+ random.shuffle(ids)
213
+ random.shuffle(ids)
214
+ random.shuffle(ids)
215
+
216
+ train, validation = np.split(ids, [int(len(ids)*0.8738)])
217
+
218
+ if split == "train":
219
+ allowed_ids = list(train)
220
+ elif split == "validation":
221
+ allowed_ids = list(validation)
222
+
223
+ for r in all_res:
224
+ if r["id"] in allowed_ids:
225
+ yield r["id"], r
226
+ else:
227
+
228
+ for r in all_res:
229
+ yield r["id"], r
test_e3c.py ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ import json
2
+
3
+ from datasets import load_dataset
4
+
5
+ dataset = load_dataset("./E3C.py", name="French")
6
+ print(dataset)
7
+ # print(dataset["train"][0])
8
+ print(json.dumps(dataset["train"][0], sort_keys=True, indent=4))