Datasets:
Tasks:
Text Classification
Sub-tasks:
text-scoring
Languages:
English
Size:
1M<n<10M
Tags:
toxicity-prediction
License:
Commit
•
700e2fa
0
Parent(s):
Update files from the datasets library (from 1.13.0)
Browse filesRelease notes: https://github.com/huggingface/datasets/releases/tag/1.13.0
- .gitattributes +27 -0
- README.md +253 -0
- dataset_infos.json +1 -0
- dummy/1.1.0/dummy_data.zip +3 -0
- jigsaw_unintended_bias.py +159 -0
.gitattributes
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
4 |
+
*.bin.* filter=lfs diff=lfs merge=lfs -text
|
5 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
11 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
12 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
13 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
14 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
15 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
16 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
17 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
18 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
19 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
20 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
21 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
22 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
23 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
24 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
25 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
26 |
+
*.zstandard filter=lfs diff=lfs merge=lfs -text
|
27 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
@@ -0,0 +1,253 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
annotations_creators:
|
3 |
+
- crowdsourced
|
4 |
+
language_creators:
|
5 |
+
- crowdsourced
|
6 |
+
languages:
|
7 |
+
- en
|
8 |
+
licenses:
|
9 |
+
- cc0-1-0
|
10 |
+
multilinguality:
|
11 |
+
- monolingual
|
12 |
+
pretty_name: Jigsaw Unintended Bias in Toxicity Classification
|
13 |
+
size_categories:
|
14 |
+
- 1M<n<10M
|
15 |
+
source_datasets:
|
16 |
+
- original
|
17 |
+
task_categories:
|
18 |
+
- text-scoring
|
19 |
+
task_ids:
|
20 |
+
- text-scoring-other-toxicity-prediction
|
21 |
+
---
|
22 |
+
|
23 |
+
# Dataset Card for Jigsaw Unintended Bias in Toxicity Classification
|
24 |
+
|
25 |
+
## Table of Contents
|
26 |
+
- [Table of Contents](#table-of-contents)
|
27 |
+
- [Dataset Description](#dataset-description)
|
28 |
+
- [Dataset Summary](#dataset-summary)
|
29 |
+
- [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
|
30 |
+
- [Languages](#languages)
|
31 |
+
- [Dataset Structure](#dataset-structure)
|
32 |
+
- [Data Instances](#data-instances)
|
33 |
+
- [Data Fields](#data-fields)
|
34 |
+
- [Data Splits](#data-splits)
|
35 |
+
- [Dataset Creation](#dataset-creation)
|
36 |
+
- [Curation Rationale](#curation-rationale)
|
37 |
+
- [Source Data](#source-data)
|
38 |
+
- [Annotations](#annotations)
|
39 |
+
- [Personal and Sensitive Information](#personal-and-sensitive-information)
|
40 |
+
- [Considerations for Using the Data](#considerations-for-using-the-data)
|
41 |
+
- [Social Impact of Dataset](#social-impact-of-dataset)
|
42 |
+
- [Discussion of Biases](#discussion-of-biases)
|
43 |
+
- [Other Known Limitations](#other-known-limitations)
|
44 |
+
- [Additional Information](#additional-information)
|
45 |
+
- [Dataset Curators](#dataset-curators)
|
46 |
+
- [Licensing Information](#licensing-information)
|
47 |
+
- [Citation Information](#citation-information)
|
48 |
+
- [Contributions](#contributions)
|
49 |
+
|
50 |
+
## Dataset Description
|
51 |
+
|
52 |
+
- **Homepage: https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification **
|
53 |
+
- **Repository: N/A **
|
54 |
+
- **Paper: N/A **
|
55 |
+
- **Leaderboard: https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/leaderboard **
|
56 |
+
- **Point of Contact: N/A **
|
57 |
+
|
58 |
+
### Dataset Summary
|
59 |
+
|
60 |
+
The Jigsaw Unintended Bias in Toxicity Classification dataset comes from the eponymous Kaggle competition.
|
61 |
+
|
62 |
+
Please see the original [data](https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/data)
|
63 |
+
description for more information.
|
64 |
+
|
65 |
+
### Supported Tasks and Leaderboards
|
66 |
+
|
67 |
+
The main target for this dataset is toxicity prediction. Several toxicity subtypes are also available, so the dataset
|
68 |
+
can be used for multi-attribute prediction.
|
69 |
+
|
70 |
+
See the original [leaderboard](https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/leaderboard)
|
71 |
+
for reference.
|
72 |
+
|
73 |
+
### Languages
|
74 |
+
|
75 |
+
English
|
76 |
+
|
77 |
+
## Dataset Structure
|
78 |
+
|
79 |
+
### Data Instances
|
80 |
+
|
81 |
+
A data point consists of an id, a comment, the main target, the other toxicity subtypes as well as identity attributes.
|
82 |
+
|
83 |
+
For instance, here's the first train example.
|
84 |
+
```
|
85 |
+
{
|
86 |
+
"article_id": 2006,
|
87 |
+
"asian": NaN,
|
88 |
+
"atheist": NaN,
|
89 |
+
"bisexual": NaN,
|
90 |
+
"black": NaN,
|
91 |
+
"buddhist": NaN,
|
92 |
+
"christian": NaN,
|
93 |
+
"comment_text": "This is so cool. It's like, 'would you want your mother to read this??' Really great idea, well done!",
|
94 |
+
"created_date": "2015-09-29 10:50:41.987077+00",
|
95 |
+
"disagree": 0,
|
96 |
+
"female": NaN,
|
97 |
+
"funny": 0,
|
98 |
+
"heterosexual": NaN,
|
99 |
+
"hindu": NaN,
|
100 |
+
"homosexual_gay_or_lesbian": NaN,
|
101 |
+
"identity_annotator_count": 0,
|
102 |
+
"identity_attack": 0.0,
|
103 |
+
"insult": 0.0,
|
104 |
+
"intellectual_or_learning_disability": NaN,
|
105 |
+
"jewish": NaN,
|
106 |
+
"latino": NaN,
|
107 |
+
"likes": 0,
|
108 |
+
"male": NaN,
|
109 |
+
"muslim": NaN,
|
110 |
+
"obscene": 0.0,
|
111 |
+
"other_disability": NaN,
|
112 |
+
"other_gender": NaN,
|
113 |
+
"other_race_or_ethnicity": NaN,
|
114 |
+
"other_religion": NaN,
|
115 |
+
"other_sexual_orientation": NaN,
|
116 |
+
"parent_id": NaN,
|
117 |
+
"physical_disability": NaN,
|
118 |
+
"psychiatric_or_mental_illness": NaN,
|
119 |
+
"publication_id": 2,
|
120 |
+
"rating": 0,
|
121 |
+
"sad": 0,
|
122 |
+
"severe_toxicity": 0.0,
|
123 |
+
"sexual_explicit": 0.0,
|
124 |
+
"target": 0.0,
|
125 |
+
"threat": 0.0,
|
126 |
+
"toxicity_annotator_count": 4,
|
127 |
+
"transgender": NaN,
|
128 |
+
"white": NaN,
|
129 |
+
"wow": 0
|
130 |
+
}
|
131 |
+
```
|
132 |
+
|
133 |
+
### Data Fields
|
134 |
+
|
135 |
+
- `id`: id of the comment
|
136 |
+
- `target`: value between 0(non-toxic) and 1(toxic) classifying the comment
|
137 |
+
- `comment_text`: the text of the comment
|
138 |
+
- `severe_toxicity`: value between 0(non-severe_toxic) and 1(severe_toxic) classifying the comment
|
139 |
+
- `obscene`: value between 0(non-obscene) and 1(obscene) classifying the comment
|
140 |
+
- `identity_attack`: value between 0(non-identity_hate) or 1(identity_hate) classifying the comment
|
141 |
+
- `insult`: value between 0(non-insult) or 1(insult) classifying the comment
|
142 |
+
- `threat`: value between 0(non-threat) and 1(threat) classifying the comment
|
143 |
+
- For a subset of rows, columns containing whether the comment mentions the entities (they may contain NaNs):
|
144 |
+
- `male`
|
145 |
+
- `female`
|
146 |
+
- `transgender`
|
147 |
+
- `other_gender`
|
148 |
+
- `heterosexual`
|
149 |
+
- `homosexual_gay_or_lesbian`
|
150 |
+
- `bisexual`
|
151 |
+
- `other_sexual_orientation`
|
152 |
+
- `christian`
|
153 |
+
- `jewish`
|
154 |
+
- `muslim`
|
155 |
+
- `hindu`
|
156 |
+
- `buddhist`
|
157 |
+
- `atheist`
|
158 |
+
- `other_religion`
|
159 |
+
- `black`
|
160 |
+
- `white`
|
161 |
+
- `asian`
|
162 |
+
- `latino`
|
163 |
+
- `other_race_or_ethnicity`
|
164 |
+
- `physical_disability`
|
165 |
+
- `intellectual_or_learning_disability`
|
166 |
+
- `psychiatric_or_mental_illness`
|
167 |
+
- `other_disability`
|
168 |
+
- Other metadata related to the source of the comment, such as creation date, publication id, number of likes,
|
169 |
+
number of annotators, etc:
|
170 |
+
- `created_date`
|
171 |
+
- `publication_id`
|
172 |
+
- `parent_id`
|
173 |
+
- `article_id`
|
174 |
+
- `rating`
|
175 |
+
- `funny`
|
176 |
+
- `wow`
|
177 |
+
- `sad`
|
178 |
+
- `likes`
|
179 |
+
- `disagree`
|
180 |
+
- `sexual_explicit`
|
181 |
+
- `identity_annotator_count`
|
182 |
+
- `toxicity_annotator_count`
|
183 |
+
|
184 |
+
### Data Splits
|
185 |
+
|
186 |
+
There are four splits:
|
187 |
+
- train: The train dataset as released during the competition. Contains labels and identity information for a
|
188 |
+
subset of rows.
|
189 |
+
- test: The train dataset as released during the competition. Does not contain labels nor identity information.
|
190 |
+
- test_private_expanded: The private leaderboard test set, including toxicity labels and subgroups. The competition target was a binarized version of the toxicity column, which can be easily reconstructed using a >=0.5 threshold.
|
191 |
+
- test_public_expanded: The public leaderboard test set, including toxicity labels and subgroups. The competition target was a binarized version of the toxicity column, which can be easily reconstructed using a >=0.5 threshold.
|
192 |
+
|
193 |
+
## Dataset Creation
|
194 |
+
|
195 |
+
### Curation Rationale
|
196 |
+
|
197 |
+
The dataset was created to help in efforts to identify and curb instances of toxicity online.
|
198 |
+
|
199 |
+
### Source Data
|
200 |
+
|
201 |
+
#### Initial Data Collection and Normalization
|
202 |
+
|
203 |
+
[More Information Needed]
|
204 |
+
|
205 |
+
#### Who are the source language producers?
|
206 |
+
|
207 |
+
[More Information Needed]
|
208 |
+
|
209 |
+
### Annotations
|
210 |
+
|
211 |
+
#### Annotation process
|
212 |
+
|
213 |
+
[More Information Needed]
|
214 |
+
|
215 |
+
#### Who are the annotators?
|
216 |
+
|
217 |
+
[More Information Needed]
|
218 |
+
|
219 |
+
### Personal and Sensitive Information
|
220 |
+
|
221 |
+
[More Information Needed]
|
222 |
+
|
223 |
+
## Considerations for Using the Data
|
224 |
+
|
225 |
+
### Social Impact of Dataset
|
226 |
+
|
227 |
+
[More Information Needed]
|
228 |
+
|
229 |
+
### Discussion of Biases
|
230 |
+
|
231 |
+
[More Information Needed]
|
232 |
+
|
233 |
+
### Other Known Limitations
|
234 |
+
|
235 |
+
[More Information Needed]
|
236 |
+
|
237 |
+
## Additional Information
|
238 |
+
|
239 |
+
### Dataset Curators
|
240 |
+
|
241 |
+
[More Information Needed]
|
242 |
+
|
243 |
+
### Licensing Information
|
244 |
+
|
245 |
+
This dataset is released under CC0, as is the underlying comment text.
|
246 |
+
|
247 |
+
### Citation Information
|
248 |
+
|
249 |
+
No citation is available for this dataset, though you may link to the [kaggle](https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification) competition
|
250 |
+
|
251 |
+
### Contributions
|
252 |
+
|
253 |
+
Thanks to [@iwontbecreative](https://github.com/iwontbecreative) for adding this dataset.
|
dataset_infos.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"default": {"description": "A collection of comments from the defunct Civil Comments platform that have been annotated for their toxicity.\n", "citation": "", "homepage": "https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/", "license": "CC0 (both the dataset and underlying text)", "features": {"target": {"dtype": "float32", "id": null, "_type": "Value"}, "comment_text": {"dtype": "string", "id": null, "_type": "Value"}, "severe_toxicity": {"dtype": "float32", "id": null, "_type": "Value"}, "obscene": {"dtype": "float32", "id": null, "_type": "Value"}, "identity_attack": {"dtype": "float32", "id": null, "_type": "Value"}, "insult": {"dtype": "float32", "id": null, "_type": "Value"}, "threat": {"dtype": "float32", "id": null, "_type": "Value"}, "asian": {"dtype": "float32", "id": null, "_type": "Value"}, "atheist": {"dtype": "float32", "id": null, "_type": "Value"}, "bisexual": {"dtype": "float32", "id": null, "_type": "Value"}, "black": {"dtype": "float32", "id": null, "_type": "Value"}, "buddhist": {"dtype": "float32", "id": null, "_type": "Value"}, "christian": {"dtype": "float32", "id": null, "_type": "Value"}, "female": {"dtype": "float32", "id": null, "_type": "Value"}, "heterosexual": {"dtype": "float32", "id": null, "_type": "Value"}, "hindu": {"dtype": "float32", "id": null, "_type": "Value"}, "homosexual_gay_or_lesbian": {"dtype": "float32", "id": null, "_type": "Value"}, "intellectual_or_learning_disability": {"dtype": "float32", "id": null, "_type": "Value"}, "jewish": {"dtype": "float32", "id": null, "_type": "Value"}, "latino": {"dtype": "float32", "id": null, "_type": "Value"}, "male": {"dtype": "float32", "id": null, "_type": "Value"}, "muslim": {"dtype": "float32", "id": null, "_type": "Value"}, "other_disability": {"dtype": "float32", "id": null, "_type": "Value"}, "other_gender": {"dtype": "float32", "id": null, "_type": "Value"}, "other_race_or_ethnicity": {"dtype": "float32", "id": null, "_type": "Value"}, "other_religion": {"dtype": "float32", "id": null, "_type": "Value"}, "other_sexual_orientation": {"dtype": "float32", "id": null, "_type": "Value"}, "physical_disability": {"dtype": "float32", "id": null, "_type": "Value"}, "psychiatric_or_mental_illness": {"dtype": "float32", "id": null, "_type": "Value"}, "transgender": {"dtype": "float32", "id": null, "_type": "Value"}, "white": {"dtype": "float32", "id": null, "_type": "Value"}, "created_date": {"dtype": "string", "id": null, "_type": "Value"}, "publication_id": {"dtype": "int32", "id": null, "_type": "Value"}, "parent_id": {"dtype": "float32", "id": null, "_type": "Value"}, "article_id": {"dtype": "int32", "id": null, "_type": "Value"}, "rating": {"num_classes": 2, "names": ["rejected", "approved"], "names_file": null, "id": null, "_type": "ClassLabel"}, "funny": {"dtype": "int32", "id": null, "_type": "Value"}, "wow": {"dtype": "int32", "id": null, "_type": "Value"}, "sad": {"dtype": "int32", "id": null, "_type": "Value"}, "likes": {"dtype": "int32", "id": null, "_type": "Value"}, "disagree": {"dtype": "int32", "id": null, "_type": "Value"}, "sexual_explicit": {"dtype": "float32", "id": null, "_type": "Value"}, "identity_annotator_count": {"dtype": "int32", "id": null, "_type": "Value"}, "toxicity_annotator_count": {"dtype": "int32", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "jigsaw_unintended_bias", "config_name": "default", "version": {"version_str": "1.1.0", "description": null, "major": 1, "minor": 1, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 914264058, "num_examples": 1804874, "dataset_name": "jigsaw_unintended_bias"}, "test_private_leaderboard": {"name": "test_private_leaderboard", "num_bytes": 49188921, "num_examples": 97320, "dataset_name": "jigsaw_unintended_bias"}, "test_public_leaderboard": {"name": "test_public_leaderboard", "num_bytes": 49442360, "num_examples": 97320, "dataset_name": "jigsaw_unintended_bias"}}, "download_checksums": {}, "download_size": 0, "post_processing_size": null, "dataset_size": 1012895339, "size_in_bytes": 1012895339}}
|
dummy/1.1.0/dummy_data.zip
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d0e82dd9c28b8ad755a09e4eaa8d9446dac6441848e9c1291ee33ca63949f851
|
3 |
+
size 3597
|
jigsaw_unintended_bias.py
ADDED
@@ -0,0 +1,159 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# coding=utf-8
|
2 |
+
# Copyright 2021 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 |
+
"""Jigsaw Unintended Bias in Toxicity Classification dataset"""
|
16 |
+
|
17 |
+
|
18 |
+
import os
|
19 |
+
|
20 |
+
import pandas as pd
|
21 |
+
|
22 |
+
import datasets
|
23 |
+
|
24 |
+
|
25 |
+
_DESCRIPTION = """\
|
26 |
+
A collection of comments from the defunct Civil Comments platform that have been annotated for their toxicity.
|
27 |
+
"""
|
28 |
+
|
29 |
+
_HOMEPAGE = "https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/"
|
30 |
+
|
31 |
+
_LICENSE = "CC0 (both the dataset and underlying text)"
|
32 |
+
|
33 |
+
|
34 |
+
class JigsawUnintendedBias(datasets.GeneratorBasedBuilder):
|
35 |
+
"""A collection of comments from the defunct Civil Comments platform that have been annotated for their toxicity."""
|
36 |
+
|
37 |
+
VERSION = datasets.Version("1.1.0")
|
38 |
+
|
39 |
+
@property
|
40 |
+
def manual_download_instructions(self):
|
41 |
+
return """\
|
42 |
+
To use jigsaw_unintended_bias you have to download it manually from Kaggle: https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/data
|
43 |
+
You can manually download the data from it's homepage or use the Kaggle CLI tool (follow the instructions here: https://www.kaggle.com/docs/api)
|
44 |
+
Please extract all files in one folder and then load the dataset with:
|
45 |
+
`datasets.load_dataset('jigsaw_unintended_bias', data_dir='/path/to/extracted/data/')`"""
|
46 |
+
|
47 |
+
def _info(self):
|
48 |
+
|
49 |
+
return datasets.DatasetInfo(
|
50 |
+
# This is the description that will appear on the datasets page.
|
51 |
+
description=_DESCRIPTION,
|
52 |
+
# This defines the different columns of the dataset and their types
|
53 |
+
features=datasets.Features(
|
54 |
+
{
|
55 |
+
"target": datasets.Value("float32"),
|
56 |
+
"comment_text": datasets.Value("string"),
|
57 |
+
"severe_toxicity": datasets.Value("float32"),
|
58 |
+
"obscene": datasets.Value("float32"),
|
59 |
+
"identity_attack": datasets.Value("float32"),
|
60 |
+
"insult": datasets.Value("float32"),
|
61 |
+
"threat": datasets.Value("float32"),
|
62 |
+
"asian": datasets.Value("float32"),
|
63 |
+
"atheist": datasets.Value("float32"),
|
64 |
+
"bisexual": datasets.Value("float32"),
|
65 |
+
"black": datasets.Value("float32"),
|
66 |
+
"buddhist": datasets.Value("float32"),
|
67 |
+
"christian": datasets.Value("float32"),
|
68 |
+
"female": datasets.Value("float32"),
|
69 |
+
"heterosexual": datasets.Value("float32"),
|
70 |
+
"hindu": datasets.Value("float32"),
|
71 |
+
"homosexual_gay_or_lesbian": datasets.Value("float32"),
|
72 |
+
"intellectual_or_learning_disability": datasets.Value("float32"),
|
73 |
+
"jewish": datasets.Value("float32"),
|
74 |
+
"latino": datasets.Value("float32"),
|
75 |
+
"male": datasets.Value("float32"),
|
76 |
+
"muslim": datasets.Value("float32"),
|
77 |
+
"other_disability": datasets.Value("float32"),
|
78 |
+
"other_gender": datasets.Value("float32"),
|
79 |
+
"other_race_or_ethnicity": datasets.Value("float32"),
|
80 |
+
"other_religion": datasets.Value("float32"),
|
81 |
+
"other_sexual_orientation": datasets.Value("float32"),
|
82 |
+
"physical_disability": datasets.Value("float32"),
|
83 |
+
"psychiatric_or_mental_illness": datasets.Value("float32"),
|
84 |
+
"transgender": datasets.Value("float32"),
|
85 |
+
"white": datasets.Value("float32"),
|
86 |
+
"created_date": datasets.Value("string"),
|
87 |
+
"publication_id": datasets.Value("int32"),
|
88 |
+
"parent_id": datasets.Value("float"),
|
89 |
+
"article_id": datasets.Value("int32"),
|
90 |
+
"rating": datasets.ClassLabel(names=["rejected", "approved"]),
|
91 |
+
"funny": datasets.Value("int32"),
|
92 |
+
"wow": datasets.Value("int32"),
|
93 |
+
"sad": datasets.Value("int32"),
|
94 |
+
"likes": datasets.Value("int32"),
|
95 |
+
"disagree": datasets.Value("int32"),
|
96 |
+
"sexual_explicit": datasets.Value("float"),
|
97 |
+
"identity_annotator_count": datasets.Value("int32"),
|
98 |
+
"toxicity_annotator_count": datasets.Value("int32"),
|
99 |
+
}
|
100 |
+
),
|
101 |
+
# If there's a common (input, target) tuple from the features,
|
102 |
+
# specify them here. They'll be used if as_supervised=True in
|
103 |
+
# builder.as_dataset.
|
104 |
+
supervised_keys=None,
|
105 |
+
# Homepage of the dataset for documentation
|
106 |
+
homepage=_HOMEPAGE,
|
107 |
+
# License for the dataset if available
|
108 |
+
license=_LICENSE,
|
109 |
+
)
|
110 |
+
|
111 |
+
def _split_generators(self, dl_manager):
|
112 |
+
"""Returns SplitGenerators."""
|
113 |
+
# This method is tasked with downloading/extracting the data and defining the splits depending on the configuration
|
114 |
+
# If several configurations are possible (listed in BUILDER_CONFIGS), the configuration selected by the user is in self.config.name
|
115 |
+
|
116 |
+
data_dir = os.path.abspath(os.path.expanduser(dl_manager.manual_dir))
|
117 |
+
|
118 |
+
if not os.path.exists(data_dir):
|
119 |
+
raise FileNotFoundError(
|
120 |
+
"{} does not exist. Make sure you insert a manual dir via `datasets.load_dataset('jigsaw_unintended_bias', data_dir=...)`. Manual download instructions: {}".format(
|
121 |
+
data_dir, self.manual_download_instructions
|
122 |
+
)
|
123 |
+
)
|
124 |
+
|
125 |
+
return [
|
126 |
+
datasets.SplitGenerator(
|
127 |
+
name=datasets.Split.TRAIN,
|
128 |
+
# These kwargs will be passed to _generate_examples
|
129 |
+
gen_kwargs={"path": os.path.join(data_dir, "train.csv"), "split": "train"},
|
130 |
+
),
|
131 |
+
datasets.SplitGenerator(
|
132 |
+
name=datasets.Split("test_private_leaderboard"),
|
133 |
+
# These kwargs will be passed to _generate_examples
|
134 |
+
gen_kwargs={"path": os.path.join(data_dir, "test_private_expanded.csv"), "split": "test"},
|
135 |
+
),
|
136 |
+
datasets.SplitGenerator(
|
137 |
+
name=datasets.Split("test_public_leaderboard"),
|
138 |
+
# These kwargs will be passed to _generate_examples
|
139 |
+
gen_kwargs={"path": os.path.join(data_dir, "test_public_expanded.csv"), "split": "test"},
|
140 |
+
),
|
141 |
+
]
|
142 |
+
|
143 |
+
def _generate_examples(self, split: str = "train", path: str = None):
|
144 |
+
"""Yields examples."""
|
145 |
+
# This method will receive as arguments the `gen_kwargs` defined in the previous `_split_generators` method.
|
146 |
+
# It is in charge of opening the given file and yielding (key, example) tuples from the dataset
|
147 |
+
# The key is not important, it's more here for legacy reason (legacy from tfds)
|
148 |
+
|
149 |
+
# Avoid loading everything into memory at once
|
150 |
+
all_data = pd.read_csv(path, chunksize=50000)
|
151 |
+
|
152 |
+
for data in all_data:
|
153 |
+
if split != "train":
|
154 |
+
data = data.rename(columns={"toxicity": "target"})
|
155 |
+
|
156 |
+
for _, row in data.iterrows():
|
157 |
+
example = row.to_dict()
|
158 |
+
ex_id = example.pop("id")
|
159 |
+
yield (ex_id, example)
|