Datasets:
Tasks:
Token Classification
Modalities:
Text
Sub-tasks:
named-entity-recognition
Languages:
Estonian
Size:
100K - 1M
ArXiv:
License:
File size: 9,315 Bytes
43aa529 |
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 |
import datasets
logger = datasets.logging.get_logger(__name__)
# Find for instance the citation on arxiv or on the dataset repo/website
_CITATION = """\
@inproceedings{hedderich2021analysing,
title={Analysing the Noise Model Error for Realistic Noisy Label Data},
author={Hedderich, Michael A and Zhu, Dawei and Klakow, Dietrich},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={35},
number={9},
pages={7675--7684},
year={2021}
}
@inproceedings{tkachenko-etal-2013-named,
title = "Named Entity Recognition in {E}stonian",
author = "Tkachenko, Alexander and Petmanson, Timo and Laur, Sven",
booktitle = "Proceedings of the 4th Biennial International Workshop on {B}alto-{S}lavic Natural Language Processing",
year = "2013",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/W13-2412",
}
"""
# You can copy an official description
_DESCRIPTION = """\
NoisyNER is a dataset for the evaluation of methods to handle noisy labels when training machine learning models.
It is from the NLP/Information Extraction domain and was created through a realistic distant supervision technique.
Some highlights and interesting aspects of the data are:
- Seven sets of labels with differing noise patterns to evaluate different noise levels on the same instances
- Full parallel clean labels available to compute upper performance bounds or study scenarios where a small amount of
gold-standard data can be leveraged
- Skewed label distribution (typical for Named Entity Recognition tasks)
- For some label sets: noise level higher than the true label probability
- Sequential dependencies between the labels
For more details on the dataset and its creation process, please refer to our publication
https://ojs.aaai.org/index.php/AAAI/article/view/16938 (published at AAAI'21).
"""
_HOMEPAGE = "https://github.com/uds-lsv/NoisyNER"
_LICENSE = "The original dataset is licensed under CC-BY-NC. We provide our noisy labels under CC-BY 4.0."
_URL = "https://huggingface.co/datasets/phuctrg/noisyner/raw/main/data"
class NoisyNER(datasets.GeneratorBasedBuilder):
"""
NoisyNER is a dataset for the evaluation of methods to handle noisy labels when training machine learning models.
"""
VERSION = datasets.Version("1.0.0")
BUILDER_CONFIGS = [
datasets.BuilderConfig(
name="estner_clean", version=VERSION, description="EstNER dataset with clean labels"
),
datasets.BuilderConfig(
name="NoisyNER_labelset1", version=VERSION,
description="NoisyNER dataset label set 1 "
"with automatic annotation via distant supervision based ANEA tool with no heuristics"
),
datasets.BuilderConfig(
name="NoisyNER_labelset2", version=VERSION,
description="NoisyNER dataset label set 2 "
"with automatic annotation via distant supervision based ANEA tool and "
"applying Estonian lemmatization to normalize the words"
),
datasets.BuilderConfig(
name="NoisyNER_labelset3", version=VERSION,
description="NoisyNER dataset label set 3 "
"with automatic annotation via distant supervision based ANEA tool and "
"splitting person entity names in the list, i.e. both first and last names can be matched "
"separately. Person names must have a minimum length of 4. Also, lemmatization"
),
datasets.BuilderConfig(
name="NoisyNER_labelset4", version=VERSION,
description="NoisyNER dataset label set 4 "
"with automatic annotation via distant supervision based ANEA tool and if entity names from "
"two different lists match the same word, location entities are preferred. "
"Also, lemmatization."
),
datasets.BuilderConfig(
name="NoisyNER_labelset5", version=VERSION,
description="NoisyNER dataset label set 5 "
"with automatic annotation via distant supervision based ANEA tool and "
"Locations preferred, lemmatization, splitting names with minimum length 4."
),
datasets.BuilderConfig(
name="NoisyNER_labelset6", version=VERSION,
description="NoisyNER dataset label set 6 "
"with automatic annotation via distant supervision based ANEA tool and "
"removing the entity names 'kohta', 'teine', 'naine' and 'mees' from the list of person names "
"(high false positive rate). Also, all of label set 5."
),
datasets.BuilderConfig(
name="NoisyNER_labelset7", version=VERSION,
description="NoisyNER dataset label set 7 "
"with automatic annotation via distant supervision based ANEA tool and using alternative, "
"alias names for organizations. Using additionally the identifiers Q82794, Q3957, Q7930989, "
"Q5119 and Q11881845 for locations and Q1572070 and Q7278 for organizations. "
"Also, all of label set 6."
),
]
DEFAULT_CONFIG_NAME = "estner_clean"
def _info(self):
features = datasets.Features(
{
"id": datasets.Value("string"),
"tokens": datasets.Sequence(datasets.Value("string")),
"lemmas": datasets.Sequence(datasets.Value("string")),
"grammar": datasets.Sequence(datasets.Value("string")),
"ner_tags": datasets.Sequence(
datasets.features.ClassLabel(
names=[
"O",
"B-PER",
"I-PER",
"B-ORG",
"I-ORG",
"B-LOC",
"I-LOC"
]
)
),
}
)
return datasets.DatasetInfo(
# This is the description that will appear on the datasets page.
description=_DESCRIPTION,
# This defines the different columns of the dataset and their types
features=features, # Here we define them above because they are different between the two configurations
# If there's a common (input, target) tuple from the features, uncomment supervised_keys line below and
# specify them. They'll be used if as_supervised=True in builder.as_dataset.
# supervised_keys=("sentence", "label"),
# Homepage of the dataset for documentation
homepage=_HOMEPAGE,
# License for the dataset if available
license=_LICENSE,
# Citation for the dataset
citation=_CITATION,
)
def _split_generators(self, dl_manager):
_URLS = {
str(datasets.Split.TRAIN): f'{_URL}/{self.config.name}_train.tsv',
str(datasets.Split.VALIDATION): f'{_URL}/{self.config.name}_dev.tsv',
str(datasets.Split.TEST): f'{_URL}/{self.config.name}_test.tsv',
}
downloaded_files = dl_manager.download_and_extract(_URLS)
return [datasets.SplitGenerator(name=i, gen_kwargs={"filepath": downloaded_files[str(i)]})
for i in [datasets.Split.TRAIN, datasets.Split.VALIDATION, datasets.Split.TEST]]
# method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
def _generate_examples(self, filepath):
logger.info("⏳ Generating examples from = %s", filepath)
# The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
with open(filepath, encoding="utf-8") as f:
guid = 0
tokens = []
lemmas = []
grammar_infos = []
ner_tags = []
for line in f:
if line in ["--", "", "\n", "--\n"]:
if tokens:
yield guid, {
"id": str(guid),
"tokens": tokens,
"lemmas": lemmas,
"grammar": grammar_infos,
"ner_tags": ner_tags,
}
guid += 1
tokens = []
lemmas = []
grammar_infos = []
ner_tags = []
else:
splits = line.split("\t")
tokens.append(splits[0])
lemmas.append(splits[1])
grammar_infos.append(splits[2])
ner_tags.append(splits[3].rstrip())
# last example
if tokens:
yield guid, {
"id": str(guid),
"tokens": tokens,
"lemmas": lemmas,
"grammar": grammar_infos,
"ner_tags": ner_tags,
}
|