Datasets:
Commit
·
88bda97
1
Parent(s):
41b694a
Add HateBR task
Browse files- portuguese_benchmark.py +176 -74
portuguese_benchmark.py
CHANGED
@@ -1,8 +1,10 @@
|
|
1 |
-
import textwrap
|
2 |
import datasets
|
3 |
from typing import Dict, List, Optional, Union
|
|
|
|
|
4 |
|
5 |
import xml.etree.ElementTree as ET
|
|
|
6 |
|
7 |
logger = datasets.logging.get_logger(__name__)
|
8 |
|
@@ -25,7 +27,7 @@ _LENERBR_KWARGS = dict(
|
|
25 |
label_classes=["ORGANIZACAO", "PESSOA", "TEMPO", "LOCAL", "LEGISLACAO", "JURISPRUDENCIA"],
|
26 |
data_urls={
|
27 |
"train": "https://raw.githubusercontent.com/peluz/lener-br/master/leNER-Br/train/train.conll",
|
28 |
-
"
|
29 |
"test": "https://raw.githubusercontent.com/peluz/lener-br/master/leNER-Br/test/test.conll",
|
30 |
},
|
31 |
citation=textwrap.dedent(
|
@@ -64,7 +66,7 @@ _ASSIN2_BASE_KWARGS = dict(
|
|
64 |
),
|
65 |
data_urls={
|
66 |
"train": "https://github.com/ruanchaves/assin/raw/master/sources/assin2-train-only.xml",
|
67 |
-
"
|
68 |
"test": "https://github.com/ruanchaves/assin/raw/master/sources/assin2-test.xml",
|
69 |
},
|
70 |
citation=textwrap.dedent(
|
@@ -92,18 +94,57 @@ _ASSIN2_STS_KWARGS = dict(
|
|
92 |
**_ASSIN2_BASE_KWARGS
|
93 |
)
|
94 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
95 |
|
96 |
class PTBenchmarkConfig(datasets.BuilderConfig):
|
97 |
"""BuilderConfig for PTBenchmark."""
|
98 |
|
99 |
def __init__(
|
100 |
self,
|
101 |
-
task_type,
|
102 |
-
data_urls,
|
103 |
-
citation,
|
104 |
-
url,
|
105 |
-
label_classes=None,
|
106 |
-
|
|
|
|
|
107 |
**kwargs,
|
108 |
):
|
109 |
"""BuilderConfig for GLUE.
|
@@ -130,57 +171,74 @@ class PTBenchmarkConfig(datasets.BuilderConfig):
|
|
130 |
self.data_urls = data_urls
|
131 |
self.citation = citation
|
132 |
self.url = url
|
133 |
-
self.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
134 |
|
135 |
-
def
|
136 |
bio_labels = ["O"]
|
137 |
for label_name in config.label_classes:
|
138 |
bio_labels.append("B-" + label_name)
|
139 |
bio_labels.append("I-" + label_name)
|
140 |
-
return datasets.
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
datasets.features.ClassLabel(names=bio_labels)
|
150 |
-
),
|
151 |
-
}
|
152 |
-
)
|
153 |
-
)
|
154 |
|
155 |
-
def
|
156 |
-
return datasets.
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
"sentence2": datasets.Value("string"),
|
165 |
-
"label": datasets.features.ClassLabel(names=config.label_classes),
|
166 |
-
}
|
167 |
-
)
|
168 |
-
)
|
169 |
|
170 |
-
def
|
171 |
-
return datasets.
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
184 |
|
185 |
def _conll_ner_generator(file_path):
|
186 |
with open(file_path, encoding="utf-8") as f:
|
@@ -247,40 +305,84 @@ class PTBenchmark(datasets.GeneratorBasedBuilder):
|
|
247 |
),
|
248 |
PTBenchmarkConfig(
|
249 |
**_ASSIN2_STS_KWARGS
|
|
|
|
|
|
|
250 |
)
|
251 |
]
|
252 |
|
253 |
def _info(self) -> datasets.DatasetInfo:
|
254 |
-
|
255 |
-
|
|
|
|
|
|
|
256 |
elif self.config.task_type == "rte":
|
257 |
-
|
258 |
elif self.config.task_type == "sts":
|
259 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
260 |
|
261 |
def _split_generators(self, dl_manager: datasets.DownloadManager):
|
262 |
-
|
263 |
-
|
264 |
-
|
265 |
-
|
266 |
-
|
267 |
-
|
268 |
-
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
|
275 |
-
|
276 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
277 |
|
278 |
def _generate_examples(
|
279 |
self,
|
280 |
-
file_path: Optional[str] = None
|
|
|
|
|
281 |
):
|
282 |
logger.info("⏳ Generating examples from = %s", file_path)
|
283 |
-
if self.config.task_type == "
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
284 |
yield from _conll_ner_generator(file_path)
|
285 |
elif self.config.task_type == "rte":
|
286 |
if "assin2" in self.config.name:
|
|
|
|
|
1 |
import datasets
|
2 |
from typing import Dict, List, Optional, Union
|
3 |
+
import json
|
4 |
+
import textwrap
|
5 |
|
6 |
import xml.etree.ElementTree as ET
|
7 |
+
import pandas as pd
|
8 |
|
9 |
logger = datasets.logging.get_logger(__name__)
|
10 |
|
|
|
27 |
label_classes=["ORGANIZACAO", "PESSOA", "TEMPO", "LOCAL", "LEGISLACAO", "JURISPRUDENCIA"],
|
28 |
data_urls={
|
29 |
"train": "https://raw.githubusercontent.com/peluz/lener-br/master/leNER-Br/train/train.conll",
|
30 |
+
"validation": "https://raw.githubusercontent.com/peluz/lener-br/master/leNER-Br/dev/dev.conll",
|
31 |
"test": "https://raw.githubusercontent.com/peluz/lener-br/master/leNER-Br/test/test.conll",
|
32 |
},
|
33 |
citation=textwrap.dedent(
|
|
|
66 |
),
|
67 |
data_urls={
|
68 |
"train": "https://github.com/ruanchaves/assin/raw/master/sources/assin2-train-only.xml",
|
69 |
+
"validation": "https://github.com/ruanchaves/assin/raw/master/sources/assin2-dev.xml",
|
70 |
"test": "https://github.com/ruanchaves/assin/raw/master/sources/assin2-test.xml",
|
71 |
},
|
72 |
citation=textwrap.dedent(
|
|
|
94 |
**_ASSIN2_BASE_KWARGS
|
95 |
)
|
96 |
|
97 |
+
# Extracted from:
|
98 |
+
# - https://huggingface.co/datasets/ruanchaves/hatebr
|
99 |
+
# - https://github.com/franciellevargas/HateBR
|
100 |
+
_HATEBR_KWARGS = dict(
|
101 |
+
name = "HateBR",
|
102 |
+
description=textwrap.dedent(
|
103 |
+
"""\
|
104 |
+
HateBR is the first large-scale expert annotated dataset of Brazilian Instagram comments for abusive language detection
|
105 |
+
on the web and social media. The HateBR was collected from Brazilian Instagram comments of politicians and manually annotated
|
106 |
+
by specialists. It is composed of 7,000 documents annotated according to three different layers: a binary classification (offensive
|
107 |
+
versus non-offensive comments), offensiveness-level (highly, moderately, and slightly offensive messages), and nine hate speech
|
108 |
+
groups (xenophobia, racism, homophobia, sexism, religious intolerance, partyism, apology for the dictatorship, antisemitism,
|
109 |
+
and fatphobia). Each comment was annotated by three different annotators and achieved high inter-annotator agreement. Furthermore,
|
110 |
+
baseline experiments were implemented reaching 85% of F1-score outperforming the current literature dataset baselines for
|
111 |
+
the Portuguese language. We hope that the proposed expert annotated dataset may foster research on hate speech detection in the
|
112 |
+
Natural Language Processing area."""
|
113 |
+
),
|
114 |
+
task_type="classification",
|
115 |
+
file_type="csv",
|
116 |
+
label_classes=[0, 1, 2, 3],
|
117 |
+
data_urls={
|
118 |
+
"train": "https://raw.githubusercontent.com/franciellevargas/HateBR/2d18c5b9410c2dfdd6d5394caa54d608857dae7c/dataset/HateBR.csv"
|
119 |
+
},
|
120 |
+
citation=textwrap.dedent(
|
121 |
+
"""\
|
122 |
+
@inproceedings{vargas2022hatebr,
|
123 |
+
title={HateBR: A Large Expert Annotated Corpus of Brazilian Instagram Comments for Offensive Language and Hate Speech Detection},
|
124 |
+
author={Vargas, Francielle and Carvalho, Isabelle and de G{\'o}es, Fabiana Rodrigues and Pardo, Thiago and Benevenuto, Fabr{\'\i}cio},
|
125 |
+
booktitle={Proceedings of the Thirteenth Language Resources and Evaluation Conference},
|
126 |
+
pages={7174--7183},
|
127 |
+
year={2022}
|
128 |
+
}"""
|
129 |
+
),
|
130 |
+
url="https://github.com/franciellevargas/HateBR",
|
131 |
+
text_and_label_columns=["instagram_comments", "offensiveness_levels"],
|
132 |
+
indexes_url="https://huggingface.co/datasets/ruanchaves/hatebr/raw/main/indexes.json"
|
133 |
+
)
|
134 |
|
135 |
class PTBenchmarkConfig(datasets.BuilderConfig):
|
136 |
"""BuilderConfig for PTBenchmark."""
|
137 |
|
138 |
def __init__(
|
139 |
self,
|
140 |
+
task_type: str,
|
141 |
+
data_urls: Dict[str, str],
|
142 |
+
citation: str,
|
143 |
+
url: str,
|
144 |
+
label_classes: Optional[List[Union[str, int]]] = None,
|
145 |
+
file_type: Optional[str] = None, #filetype (csv, tsc, jsonl)
|
146 |
+
text_and_label_columns: Optional[List[str]] = None, #columns for train, dev and test for csv datasets
|
147 |
+
indexes_url=None, #indexes for train, dev and test for single file datasets
|
148 |
**kwargs,
|
149 |
):
|
150 |
"""BuilderConfig for GLUE.
|
|
|
171 |
self.data_urls = data_urls
|
172 |
self.citation = citation
|
173 |
self.url = url
|
174 |
+
self.file_type = file_type
|
175 |
+
self.text_and_label_columns = text_and_label_columns
|
176 |
+
self.indexes_url = indexes_url
|
177 |
+
|
178 |
+
def _get_classification_features(config):
|
179 |
+
return datasets.Features(
|
180 |
+
{
|
181 |
+
"idx": datasets.Value("int32"),
|
182 |
+
"sentence": datasets.Value("string"),
|
183 |
+
"label": datasets.features.ClassLabel(names=config.label_classes),
|
184 |
+
}
|
185 |
+
)
|
186 |
|
187 |
+
def _get_ner_features(config):
|
188 |
bio_labels = ["O"]
|
189 |
for label_name in config.label_classes:
|
190 |
bio_labels.append("B-" + label_name)
|
191 |
bio_labels.append("I-" + label_name)
|
192 |
+
return datasets.Features(
|
193 |
+
{
|
194 |
+
"idx": datasets.Value("int32"),
|
195 |
+
"tokens": datasets.Sequence(datasets.Value("string")),
|
196 |
+
"ner_tags": datasets.Sequence(
|
197 |
+
datasets.features.ClassLabel(names=bio_labels)
|
198 |
+
),
|
199 |
+
}
|
200 |
+
)
|
|
|
|
|
|
|
|
|
|
|
201 |
|
202 |
+
def _get_rte_features(config):
|
203 |
+
return datasets.Features(
|
204 |
+
{
|
205 |
+
"idx": datasets.Value("int32"),
|
206 |
+
"sentence1": datasets.Value("string"),
|
207 |
+
"sentence2": datasets.Value("string"),
|
208 |
+
"label": datasets.features.ClassLabel(names=config.label_classes),
|
209 |
+
}
|
210 |
+
)
|
|
|
|
|
|
|
|
|
|
|
211 |
|
212 |
+
def _get_sts_features(config):
|
213 |
+
return datasets.Features(
|
214 |
+
{
|
215 |
+
"idx": datasets.Value("int32"),
|
216 |
+
"sentence1": datasets.Value("string"),
|
217 |
+
"sentence2": datasets.Value("string"),
|
218 |
+
"label": datasets.Value("float32"),
|
219 |
+
}
|
220 |
+
)
|
221 |
+
|
222 |
+
def _csv_generator(file_path: str,
|
223 |
+
columns: List[str],
|
224 |
+
indexes_path: Optional[str] = None,
|
225 |
+
split: Optional[str] = None):
|
226 |
+
"""Yields examples."""
|
227 |
+
df = pd.read_csv(file_path)
|
228 |
+
df = df[columns]
|
229 |
+
|
230 |
+
with open(indexes_path, "r") as f:
|
231 |
+
indexes= json.load(f)
|
232 |
+
split_indexes = indexes[split]
|
233 |
+
df_split = df.iloc[split_indexes]
|
234 |
+
|
235 |
+
for id_, row in df_split.iterrows():
|
236 |
+
example = {
|
237 |
+
"idx": id_,
|
238 |
+
"sentence": str(row[columns[0]]),
|
239 |
+
"label": int(row[columns[-1]])
|
240 |
+
}
|
241 |
+
yield id_, example
|
242 |
|
243 |
def _conll_ner_generator(file_path):
|
244 |
with open(file_path, encoding="utf-8") as f:
|
|
|
305 |
),
|
306 |
PTBenchmarkConfig(
|
307 |
**_ASSIN2_STS_KWARGS
|
308 |
+
),
|
309 |
+
PTBenchmarkConfig(
|
310 |
+
**_HATEBR_KWARGS
|
311 |
)
|
312 |
]
|
313 |
|
314 |
def _info(self) -> datasets.DatasetInfo:
|
315 |
+
features = None
|
316 |
+
if self.config.task_type == "classification":
|
317 |
+
features = _get_classification_features(self.config)
|
318 |
+
elif self.config.task_type == "ner":
|
319 |
+
features = _get_ner_features(self.config)
|
320 |
elif self.config.task_type == "rte":
|
321 |
+
features = _get_rte_features(self.config)
|
322 |
elif self.config.task_type == "sts":
|
323 |
+
features = _get_sts_features(self.config)
|
324 |
+
|
325 |
+
return datasets.DatasetInfo(
|
326 |
+
description=self.config.description,
|
327 |
+
homepage=self.config.url,
|
328 |
+
citation=self.config.citation,
|
329 |
+
supervised_keys=None,
|
330 |
+
features=features
|
331 |
+
)
|
332 |
|
333 |
def _split_generators(self, dl_manager: datasets.DownloadManager):
|
334 |
+
data_urls = self.config.data_urls.copy()
|
335 |
+
if self.config.indexes_url is not None:
|
336 |
+
data_urls['indexes'] = self.config.indexes_url
|
337 |
+
file_paths = dl_manager.download_and_extract(data_urls)
|
338 |
+
|
339 |
+
if self.config.indexes_url is None:
|
340 |
+
return [
|
341 |
+
datasets.SplitGenerator(
|
342 |
+
name=datasets.Split.TRAIN,
|
343 |
+
gen_kwargs={"file_path": file_paths["train"]},
|
344 |
+
),
|
345 |
+
datasets.SplitGenerator(
|
346 |
+
name=datasets.Split.VALIDATION,
|
347 |
+
gen_kwargs={"file_path": file_paths["validation"]},
|
348 |
+
),
|
349 |
+
datasets.SplitGenerator(
|
350 |
+
name=datasets.Split.TEST,
|
351 |
+
gen_kwargs={"file_path": file_paths["test"]},
|
352 |
+
)
|
353 |
+
]
|
354 |
+
else:
|
355 |
+
return [
|
356 |
+
datasets.SplitGenerator(
|
357 |
+
name=datasets.Split.TRAIN,
|
358 |
+
gen_kwargs={"file_path": file_paths["train"], "indexes_path": file_paths["indexes"], "split": "train"},
|
359 |
+
),
|
360 |
+
datasets.SplitGenerator(
|
361 |
+
name=datasets.Split.VALIDATION,
|
362 |
+
gen_kwargs={"file_path": file_paths["train"], "indexes_path": file_paths["indexes"], "split": "validation"},
|
363 |
+
),
|
364 |
+
datasets.SplitGenerator(
|
365 |
+
name=datasets.Split.TEST,
|
366 |
+
gen_kwargs={"file_path": file_paths["train"], "indexes_path": file_paths["indexes"], "split": "test"},
|
367 |
+
)
|
368 |
+
]
|
369 |
|
370 |
def _generate_examples(
|
371 |
self,
|
372 |
+
file_path: Optional[str] = None,
|
373 |
+
indexes_path: Optional[str] = None,
|
374 |
+
split: Optional[str] = None
|
375 |
):
|
376 |
logger.info("⏳ Generating examples from = %s", file_path)
|
377 |
+
if self.config.task_type == "classification":
|
378 |
+
if self.config.file_type == "csv":
|
379 |
+
yield from _csv_generator(
|
380 |
+
file_path,
|
381 |
+
self.config.text_and_label_columns,
|
382 |
+
indexes_path=indexes_path,
|
383 |
+
split=split
|
384 |
+
)
|
385 |
+
elif self.config.task_type == "ner":
|
386 |
yield from _conll_ner_generator(file_path)
|
387 |
elif self.config.task_type == "rte":
|
388 |
if "assin2" in self.config.name:
|