Datasets:
Tasks:
Token Classification
Modalities:
Text
Sub-tasks:
named-entity-recognition
Languages:
Estonian
Size:
100K - 1M
ArXiv:
License:
Add data files, loading script and update README.md
Browse files- README.md +218 -1
- data/NoisyNER_labelset1_all.tsv +0 -0
- data/NoisyNER_labelset1_dev.tsv +0 -0
- data/NoisyNER_labelset1_test.tsv +0 -0
- data/NoisyNER_labelset1_train.tsv +0 -0
- data/NoisyNER_labelset2_all.tsv +0 -0
- data/NoisyNER_labelset2_dev.tsv +0 -0
- data/NoisyNER_labelset2_test.tsv +0 -0
- data/NoisyNER_labelset2_train.tsv +0 -0
- data/NoisyNER_labelset3_all.tsv +0 -0
- data/NoisyNER_labelset3_dev.tsv +0 -0
- data/NoisyNER_labelset3_test.tsv +0 -0
- data/NoisyNER_labelset3_train.tsv +0 -0
- data/NoisyNER_labelset4_all.tsv +0 -0
- data/NoisyNER_labelset4_dev.tsv +0 -0
- data/NoisyNER_labelset4_test.tsv +0 -0
- data/NoisyNER_labelset4_train.tsv +0 -0
- data/NoisyNER_labelset5_all.tsv +0 -0
- data/NoisyNER_labelset5_dev.tsv +0 -0
- data/NoisyNER_labelset5_test.tsv +0 -0
- data/NoisyNER_labelset5_train.tsv +0 -0
- data/NoisyNER_labelset6_all.tsv +0 -0
- data/NoisyNER_labelset6_dev.tsv +0 -0
- data/NoisyNER_labelset6_test.tsv +0 -0
- data/NoisyNER_labelset6_train.tsv +0 -0
- data/NoisyNER_labelset7_all.tsv +0 -0
- data/NoisyNER_labelset7_dev.tsv +0 -0
- data/NoisyNER_labelset7_test.tsv +0 -0
- data/NoisyNER_labelset7_train.tsv +0 -0
- data/estner_clean_dev.tsv +0 -0
- data/estner_clean_test.tsv +0 -0
- data/estner_clean_train.tsv +0 -0
- noisyner.py +199 -0
README.md
CHANGED
@@ -1,3 +1,220 @@
|
|
1 |
---
|
2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
annotations_creators:
|
3 |
+
- expert-generated
|
4 |
+
language:
|
5 |
+
- et
|
6 |
+
language_creators:
|
7 |
+
- found
|
8 |
+
license:
|
9 |
+
- cc-by-nc-4.0
|
10 |
+
multilinguality:
|
11 |
+
- monolingual
|
12 |
+
paperswithcode_id: noisyner
|
13 |
+
pretty_name: NoisyNER
|
14 |
+
size_categories:
|
15 |
+
- 10K<n<100K
|
16 |
+
source_datasets:
|
17 |
+
- original
|
18 |
+
tags:
|
19 |
+
- newspapers
|
20 |
+
- 1997-2009
|
21 |
+
task_categories:
|
22 |
+
- token-classification
|
23 |
+
task_ids:
|
24 |
+
- named-entity-recognition
|
25 |
---
|
26 |
+
|
27 |
+
# Dataset Card for NoisyNER
|
28 |
+
|
29 |
+
## Table of Contents
|
30 |
+
- [Table of Contents](#table-of-contents)
|
31 |
+
- [Dataset Description](#dataset-description)
|
32 |
+
- [Dataset Summary](#dataset-summary)
|
33 |
+
- [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
|
34 |
+
- [Languages](#languages)
|
35 |
+
- [Dataset Structure](#dataset-structure)
|
36 |
+
- [Data Instances](#data-instances)
|
37 |
+
- [Data Fields](#data-fields)
|
38 |
+
- [Data Splits](#data-splits)
|
39 |
+
- [Dataset Creation](#dataset-creation)
|
40 |
+
- [Curation Rationale](#curation-rationale)
|
41 |
+
- [Source Data](#source-data)
|
42 |
+
- [Annotations](#annotations)
|
43 |
+
- [Personal and Sensitive Information](#personal-and-sensitive-information)
|
44 |
+
- [Considerations for Using the Data](#considerations-for-using-the-data)
|
45 |
+
- [Social Impact of Dataset](#social-impact-of-dataset)
|
46 |
+
- [Discussion of Biases](#discussion-of-biases)
|
47 |
+
- [Other Known Limitations](#other-known-limitations)
|
48 |
+
- [Additional Information](#additional-information)
|
49 |
+
- [Dataset Curators](#dataset-curators)
|
50 |
+
- [Licensing Information](#licensing-information)
|
51 |
+
- [Citation Information](#citation-information)
|
52 |
+
- [Contributions](#contributions)
|
53 |
+
|
54 |
+
## Dataset Description
|
55 |
+
|
56 |
+
- **Repository:** [Estonian NER corpus](https://doi.org/10.15155/1-00-0000-0000-0000-00073L), [NoisyNER dataset](https://github.com/uds-lsv/NoisyNER)
|
57 |
+
- **Paper:** [Named Entity Recognition in Estonian](https://aclanthology.org/W13-2412/), [Analysing the Noise Model Error for Realistic Noisy Label Data](https://ojs.aaai.org/index.php/AAAI/article/view/16938)
|
58 |
+
- **Dataset:** NoisyNER
|
59 |
+
- **Domain:** News
|
60 |
+
|
61 |
+
### Dataset Summary
|
62 |
+
|
63 |
+
NoisyNER is a dataset for the evaluation of methods to handle noisy labels when training machine learning models.
|
64 |
+
|
65 |
+
- Entity Types: `PER`, `ORG`, `LOC`
|
66 |
+
|
67 |
+
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:
|
68 |
+
|
69 |
+
- Seven sets of labels with differing noise patterns to evaluate different noise levels on the same instances
|
70 |
+
- Full parallel clean labels available to compute upper performance bounds or study scenarios where a small amount of gold-standard data can be leveraged
|
71 |
+
- Skewed label distribution (typical for Named Entity Recognition tasks)
|
72 |
+
- For some label sets: noise level higher than the true label probability
|
73 |
+
- Sequential dependencies between the labels
|
74 |
+
|
75 |
+
For more details on the dataset and its creation process, please refer to the original author's publication https://ojs.aaai.org/index.php/AAAI/article/view/16938 (published at AAAI'21).
|
76 |
+
|
77 |
+
This dataset is based on the Estonian NER corpus. For more details see https://aclanthology.org/W13-2412/
|
78 |
+
|
79 |
+
### Supported Tasks and Leaderboards
|
80 |
+
|
81 |
+
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
82 |
+
|
83 |
+
### Languages
|
84 |
+
|
85 |
+
The language data in NoisyNER is in Estonian (BCP-47 et)
|
86 |
+
|
87 |
+
## Dataset Structure
|
88 |
+
|
89 |
+
### Data Instances
|
90 |
+
|
91 |
+
An example of 'train' looks as follows.
|
92 |
+
```
|
93 |
+
{
|
94 |
+
'id': '0',
|
95 |
+
'tokens': ['Tallinna', 'õhusaaste', 'suureneb', '.'],
|
96 |
+
'lemmas': ['Tallinn+0', 'õhu_saaste+0', 'suurene+b', '.'],
|
97 |
+
'grammar': ['_H_ sg g', '_S_ sg n', '_V_ b', '_Z_'],
|
98 |
+
'ner_tags': [5, 0, 0, 0]
|
99 |
+
}
|
100 |
+
```
|
101 |
+
|
102 |
+
### Data Fields
|
103 |
+
|
104 |
+
The data fields are the same among all splits.
|
105 |
+
|
106 |
+
- `id`: a `string` feature.
|
107 |
+
- `tokens`: a `list` of `string` features.
|
108 |
+
- `lemmas`: a `list` of `string` features.
|
109 |
+
- `grammar`: a `list` of `string` features.
|
110 |
+
- `ner_tags`: a `list` of classification labels (`int`). Full tagset with indices:
|
111 |
+
|
112 |
+
```python
|
113 |
+
{'O': 0, 'B-PER': 1, 'I-PER': 2, 'B-ORG': 3, 'I-ORG': 4, 'B-LOC': 5, 'I-LOC': 6}
|
114 |
+
```
|
115 |
+
|
116 |
+
### Data Splits
|
117 |
+
|
118 |
+
The splits are the same across all configurations.
|
119 |
+
|
120 |
+
|train|validation|test|
|
121 |
+
|----:|---------:|---:|
|
122 |
+
|11365| 1480|1433|
|
123 |
+
|
124 |
+
## Dataset Creation
|
125 |
+
|
126 |
+
### Curation Rationale
|
127 |
+
|
128 |
+
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
129 |
+
|
130 |
+
### Source Data
|
131 |
+
|
132 |
+
#### Initial Data Collection and Normalization
|
133 |
+
|
134 |
+
Tkachenko et al (2013) collected 572 news stories published in the local online newspapers [Delfi](http://delfi.ee/) and [Postimees](http://postimees.ee/) between 1997 and 2009. Selected articles cover both local and international news on a range of topics including politics, economics and sports. The raw text was preprocessed using the morphological disambiguator t3mesta ([Kaalep and
|
135 |
+
Vaino, 1998](https://www.cl.ut.ee/yllitised/kk_yhest_1998.pdf)) provided by [Filosoft](http://www.filosoft.ee/). The processing steps involve tokenization, lemmatization, part-of-speech tagging, grammatical and morphological analysis.
|
136 |
+
|
137 |
+
#### Who are the source language producers?
|
138 |
+
|
139 |
+
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
140 |
+
|
141 |
+
### Annotations
|
142 |
+
|
143 |
+
#### Annotation process
|
144 |
+
|
145 |
+
According to Tkachenko et al (2013) one of the authors manually tagged the corpus and the other author examined the tags, after which conflicting cases were resolved.
|
146 |
+
The total size of the corpus is 184,638 tokens. Tkachenko et al (2013) provide the following number of named entities in the corpus:
|
147 |
+
|
148 |
+
| | PER | LOC | ORG | Total |
|
149 |
+
|--------|------|------|------|-------|
|
150 |
+
| All | 5762 | 5711 | 3938 | 15411 |
|
151 |
+
| Unique | 3588 | 1589 | 1987 | 7164 |
|
152 |
+
|
153 |
+
Hedderich et al (2021) obtained the noisy labels through a distant supervision/automatic annotation approach. They extracted lists of named entities from Wikidata and matched them against words in the text via the ANEA tool ([Hedderich, Lange, and Klakow 2021](https://arxiv.org/abs/2102.13129)). They also used heuristic functions to correct errors caused by non-complete lists of entities,
|
154 |
+
grammatical complexities of Estonian that do not allow simple string matching or entity lists in conflict with each other. For instance, they normalized the grammatical form of a word or excluded certain high false-positive words. They provide seven sets of labels that differ in the noise process. This results in 8 different configurations, when added to the original split with clean labels.
|
155 |
+
|
156 |
+
#### Who are the annotators?
|
157 |
+
|
158 |
+
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
159 |
+
|
160 |
+
### Personal and Sensitive Information
|
161 |
+
|
162 |
+
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
163 |
+
|
164 |
+
## Considerations for Using the Data
|
165 |
+
|
166 |
+
### Social Impact of Dataset
|
167 |
+
|
168 |
+
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
169 |
+
|
170 |
+
### Discussion of Biases
|
171 |
+
|
172 |
+
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
173 |
+
|
174 |
+
### Other Known Limitations
|
175 |
+
|
176 |
+
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
177 |
+
|
178 |
+
## Additional Information
|
179 |
+
|
180 |
+
### Dataset Curators
|
181 |
+
|
182 |
+
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
183 |
+
|
184 |
+
### Licensing Information
|
185 |
+
|
186 |
+
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
187 |
+
|
188 |
+
|
189 |
+
### Citation Information
|
190 |
+
|
191 |
+
```
|
192 |
+
@inproceedings{tkachenko-etal-2013-named,
|
193 |
+
title = "Named Entity Recognition in {E}stonian",
|
194 |
+
author = "Tkachenko, Alexander and
|
195 |
+
Petmanson, Timo and
|
196 |
+
Laur, Sven",
|
197 |
+
booktitle = "Proceedings of the 4th Biennial International Workshop on {B}alto-{S}lavic Natural Language Processing",
|
198 |
+
month = aug,
|
199 |
+
year = "2013",
|
200 |
+
address = "Sofia, Bulgaria",
|
201 |
+
publisher = "Association for Computational Linguistics",
|
202 |
+
url = "https://aclanthology.org/W13-2412",
|
203 |
+
pages = "78--83",
|
204 |
+
}
|
205 |
+
@article{Hedderich_Zhu_Klakow_2021,
|
206 |
+
title={Analysing the Noise Model Error for Realistic Noisy Label Data},
|
207 |
+
author={Hedderich, Michael A. and Zhu, Dawei and Klakow, Dietrich},
|
208 |
+
volume={35},
|
209 |
+
url={https://ojs.aaai.org/index.php/AAAI/article/view/16938},
|
210 |
+
number={9},
|
211 |
+
journal={Proceedings of the AAAI Conference on Artificial Intelligence},
|
212 |
+
year={2021},
|
213 |
+
month={May},
|
214 |
+
pages={7675-7684},
|
215 |
+
}
|
216 |
+
```
|
217 |
+
|
218 |
+
### Contributions
|
219 |
+
|
220 |
+
Thanks to [@phucdev](https://github.com/phucdev) for adding this dataset.
|
data/NoisyNER_labelset1_all.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset1_dev.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset1_test.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset1_train.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset2_all.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset2_dev.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset2_test.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset2_train.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset3_all.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset3_dev.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset3_test.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset3_train.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset4_all.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset4_dev.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset4_test.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset4_train.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset5_all.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset5_dev.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset5_test.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset5_train.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset6_all.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset6_dev.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset6_test.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset6_train.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset7_all.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset7_dev.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset7_test.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/NoisyNER_labelset7_train.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/estner_clean_dev.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/estner_clean_test.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/estner_clean_train.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
noisyner.py
ADDED
@@ -0,0 +1,199 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import datasets
|
2 |
+
|
3 |
+
|
4 |
+
logger = datasets.logging.get_logger(__name__)
|
5 |
+
|
6 |
+
# Find for instance the citation on arxiv or on the dataset repo/website
|
7 |
+
_CITATION = """\
|
8 |
+
@inproceedings{hedderich2021analysing,
|
9 |
+
title={Analysing the Noise Model Error for Realistic Noisy Label Data},
|
10 |
+
author={Hedderich, Michael A and Zhu, Dawei and Klakow, Dietrich},
|
11 |
+
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
|
12 |
+
volume={35},
|
13 |
+
number={9},
|
14 |
+
pages={7675--7684},
|
15 |
+
year={2021}
|
16 |
+
}
|
17 |
+
@inproceedings{tkachenko-etal-2013-named,
|
18 |
+
title = "Named Entity Recognition in {E}stonian",
|
19 |
+
author = "Tkachenko, Alexander and Petmanson, Timo and Laur, Sven",
|
20 |
+
booktitle = "Proceedings of the 4th Biennial International Workshop on {B}alto-{S}lavic Natural Language Processing",
|
21 |
+
year = "2013",
|
22 |
+
publisher = "Association for Computational Linguistics",
|
23 |
+
url = "https://www.aclweb.org/anthology/W13-2412",
|
24 |
+
}
|
25 |
+
"""
|
26 |
+
|
27 |
+
# You can copy an official description
|
28 |
+
_DESCRIPTION = """\
|
29 |
+
NoisyNER is a dataset for the evaluation of methods to handle noisy labels when training machine learning models.
|
30 |
+
It is from the NLP/Information Extraction domain and was created through a realistic distant supervision technique.
|
31 |
+
Some highlights and interesting aspects of the data are:
|
32 |
+
- Seven sets of labels with differing noise patterns to evaluate different noise levels on the same instances
|
33 |
+
- Full parallel clean labels available to compute upper performance bounds or study scenarios where a small amount of
|
34 |
+
gold-standard data can be leveraged
|
35 |
+
- Skewed label distribution (typical for Named Entity Recognition tasks)
|
36 |
+
- For some label sets: noise level higher than the true label probability
|
37 |
+
- Sequential dependencies between the labels
|
38 |
+
|
39 |
+
For more details on the dataset and its creation process, please refer to our publication
|
40 |
+
https://ojs.aaai.org/index.php/AAAI/article/view/16938 (published at AAAI'21).
|
41 |
+
"""
|
42 |
+
|
43 |
+
_HOMEPAGE = "https://github.com/uds-lsv/NoisyNER"
|
44 |
+
|
45 |
+
_LICENSE = "The original dataset is licensed under CC-BY-NC. We provide our noisy labels under CC-BY 4.0."
|
46 |
+
|
47 |
+
_URL = "https://huggingface.co/datasets/phuctrg/noisyner/raw/main/data"
|
48 |
+
|
49 |
+
|
50 |
+
class NoisyNER(datasets.GeneratorBasedBuilder):
|
51 |
+
"""
|
52 |
+
NoisyNER is a dataset for the evaluation of methods to handle noisy labels when training machine learning models.
|
53 |
+
"""
|
54 |
+
|
55 |
+
VERSION = datasets.Version("1.0.0")
|
56 |
+
BUILDER_CONFIGS = [
|
57 |
+
datasets.BuilderConfig(
|
58 |
+
name="estner_clean", version=VERSION, description="EstNER dataset with clean labels"
|
59 |
+
),
|
60 |
+
datasets.BuilderConfig(
|
61 |
+
name="NoisyNER_labelset1", version=VERSION,
|
62 |
+
description="NoisyNER dataset label set 1 "
|
63 |
+
"with automatic annotation via distant supervision based ANEA tool with no heuristics"
|
64 |
+
),
|
65 |
+
datasets.BuilderConfig(
|
66 |
+
name="NoisyNER_labelset2", version=VERSION,
|
67 |
+
description="NoisyNER dataset label set 2 "
|
68 |
+
"with automatic annotation via distant supervision based ANEA tool and "
|
69 |
+
"applying Estonian lemmatization to normalize the words"
|
70 |
+
),
|
71 |
+
datasets.BuilderConfig(
|
72 |
+
name="NoisyNER_labelset3", version=VERSION,
|
73 |
+
description="NoisyNER dataset label set 3 "
|
74 |
+
"with automatic annotation via distant supervision based ANEA tool and "
|
75 |
+
"splitting person entity names in the list, i.e. both first and last names can be matched "
|
76 |
+
"separately. Person names must have a minimum length of 4. Also, lemmatization"
|
77 |
+
),
|
78 |
+
datasets.BuilderConfig(
|
79 |
+
name="NoisyNER_labelset4", version=VERSION,
|
80 |
+
description="NoisyNER dataset label set 4 "
|
81 |
+
"with automatic annotation via distant supervision based ANEA tool and if entity names from "
|
82 |
+
"two different lists match the same word, location entities are preferred. "
|
83 |
+
"Also, lemmatization."
|
84 |
+
),
|
85 |
+
datasets.BuilderConfig(
|
86 |
+
name="NoisyNER_labelset5", version=VERSION,
|
87 |
+
description="NoisyNER dataset label set 5 "
|
88 |
+
"with automatic annotation via distant supervision based ANEA tool and "
|
89 |
+
"Locations preferred, lemmatization, splitting names with minimum length 4."
|
90 |
+
),
|
91 |
+
datasets.BuilderConfig(
|
92 |
+
name="NoisyNER_labelset6", version=VERSION,
|
93 |
+
description="NoisyNER dataset label set 6 "
|
94 |
+
"with automatic annotation via distant supervision based ANEA tool and "
|
95 |
+
"removing the entity names 'kohta', 'teine', 'naine' and 'mees' from the list of person names "
|
96 |
+
"(high false positive rate). Also, all of label set 5."
|
97 |
+
),
|
98 |
+
datasets.BuilderConfig(
|
99 |
+
name="NoisyNER_labelset7", version=VERSION,
|
100 |
+
description="NoisyNER dataset label set 7 "
|
101 |
+
"with automatic annotation via distant supervision based ANEA tool and using alternative, "
|
102 |
+
"alias names for organizations. Using additionally the identifiers Q82794, Q3957, Q7930989, "
|
103 |
+
"Q5119 and Q11881845 for locations and Q1572070 and Q7278 for organizations. "
|
104 |
+
"Also, all of label set 6."
|
105 |
+
),
|
106 |
+
]
|
107 |
+
|
108 |
+
DEFAULT_CONFIG_NAME = "estner_clean"
|
109 |
+
|
110 |
+
def _info(self):
|
111 |
+
features = datasets.Features(
|
112 |
+
{
|
113 |
+
"id": datasets.Value("string"),
|
114 |
+
"tokens": datasets.Sequence(datasets.Value("string")),
|
115 |
+
"lemmas": datasets.Sequence(datasets.Value("string")),
|
116 |
+
"grammar": datasets.Sequence(datasets.Value("string")),
|
117 |
+
"ner_tags": datasets.Sequence(
|
118 |
+
datasets.features.ClassLabel(
|
119 |
+
names=[
|
120 |
+
"O",
|
121 |
+
"B-PER",
|
122 |
+
"I-PER",
|
123 |
+
"B-ORG",
|
124 |
+
"I-ORG",
|
125 |
+
"B-LOC",
|
126 |
+
"I-LOC"
|
127 |
+
]
|
128 |
+
)
|
129 |
+
),
|
130 |
+
}
|
131 |
+
)
|
132 |
+
return datasets.DatasetInfo(
|
133 |
+
# This is the description that will appear on the datasets page.
|
134 |
+
description=_DESCRIPTION,
|
135 |
+
# This defines the different columns of the dataset and their types
|
136 |
+
features=features, # Here we define them above because they are different between the two configurations
|
137 |
+
# If there's a common (input, target) tuple from the features, uncomment supervised_keys line below and
|
138 |
+
# specify them. They'll be used if as_supervised=True in builder.as_dataset.
|
139 |
+
# supervised_keys=("sentence", "label"),
|
140 |
+
# Homepage of the dataset for documentation
|
141 |
+
homepage=_HOMEPAGE,
|
142 |
+
# License for the dataset if available
|
143 |
+
license=_LICENSE,
|
144 |
+
# Citation for the dataset
|
145 |
+
citation=_CITATION,
|
146 |
+
)
|
147 |
+
|
148 |
+
def _split_generators(self, dl_manager):
|
149 |
+
_URLS = {
|
150 |
+
str(datasets.Split.TRAIN): f'{_URL}/{self.config.name}_train.tsv',
|
151 |
+
str(datasets.Split.VALIDATION): f'{_URL}/{self.config.name}_dev.tsv',
|
152 |
+
str(datasets.Split.TEST): f'{_URL}/{self.config.name}_test.tsv',
|
153 |
+
}
|
154 |
+
|
155 |
+
downloaded_files = dl_manager.download_and_extract(_URLS)
|
156 |
+
|
157 |
+
return [datasets.SplitGenerator(name=i, gen_kwargs={"filepath": downloaded_files[str(i)]})
|
158 |
+
for i in [datasets.Split.TRAIN, datasets.Split.VALIDATION, datasets.Split.TEST]]
|
159 |
+
|
160 |
+
# method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
|
161 |
+
def _generate_examples(self, filepath):
|
162 |
+
logger.info("⏳ Generating examples from = %s", filepath)
|
163 |
+
# The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
|
164 |
+
with open(filepath, encoding="utf-8") as f:
|
165 |
+
guid = 0
|
166 |
+
tokens = []
|
167 |
+
lemmas = []
|
168 |
+
grammar_infos = []
|
169 |
+
ner_tags = []
|
170 |
+
for line in f:
|
171 |
+
if line in ["--", "", "\n", "--\n"]:
|
172 |
+
if tokens:
|
173 |
+
yield guid, {
|
174 |
+
"id": str(guid),
|
175 |
+
"tokens": tokens,
|
176 |
+
"lemmas": lemmas,
|
177 |
+
"grammar": grammar_infos,
|
178 |
+
"ner_tags": ner_tags,
|
179 |
+
}
|
180 |
+
guid += 1
|
181 |
+
tokens = []
|
182 |
+
lemmas = []
|
183 |
+
grammar_infos = []
|
184 |
+
ner_tags = []
|
185 |
+
else:
|
186 |
+
splits = line.split("\t")
|
187 |
+
tokens.append(splits[0])
|
188 |
+
lemmas.append(splits[1])
|
189 |
+
grammar_infos.append(splits[2])
|
190 |
+
ner_tags.append(splits[3].rstrip())
|
191 |
+
# last example
|
192 |
+
if tokens:
|
193 |
+
yield guid, {
|
194 |
+
"id": str(guid),
|
195 |
+
"tokens": tokens,
|
196 |
+
"lemmas": lemmas,
|
197 |
+
"grammar": grammar_infos,
|
198 |
+
"ner_tags": ner_tags,
|
199 |
+
}
|