BramVanroy commited on
Commit
3d22d5c
·
verified ·
1 Parent(s): 35376bb

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +219 -3
README.md CHANGED
@@ -1,3 +1,219 @@
1
- ---
2
- license: cc
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc
3
+ task_categories:
4
+ - text-generation
5
+ task_ids:
6
+ - language-modeling
7
+ pretty_name: ©️ Common Crawl Creative Commons
8
+ language:
9
+ - afr
10
+ - deu
11
+ - eng
12
+ - fra
13
+ - fry
14
+ - ita
15
+ - nld
16
+ - spa
17
+ configs:
18
+ - config_name: CC-MAIN-2024-51
19
+ data_files: data/CC-MAIN-2024-51/**/*.parquet
20
+ - config_name: CC-MAIN-2024-51-af
21
+ data_files:
22
+ - split: train
23
+ path: data/CC-MAIN-2024-51/af/train-*
24
+ - config_name: CC-MAIN-2024-51-de
25
+ data_files: data/CC-MAIN-2024-51/de/*.parquet
26
+ - config_name: CC-MAIN-2024-51-en
27
+ data_files: data/CC-MAIN-2024-51/en/*.parquet
28
+ - config_name: CC-MAIN-2024-51-es
29
+ data_files: data/CC-MAIN-2024-51/es/*.parquet
30
+ - config_name: CC-MAIN-2024-51-fr
31
+ data_files: data/CC-MAIN-2024-51/fr/*.parquet
32
+ - config_name: CC-MAIN-2024-51-fy
33
+ data_files: data/CC-MAIN-2024-51/fy/*.parquet
34
+ - config_name: CC-MAIN-2024-51-it
35
+ data_files: data/CC-MAIN-2024-51/it/*.parquet
36
+ - config_name: CC-MAIN-2024-51-nl
37
+ data_files: data/CC-MAIN-2024-51/nl/*.parquet
38
+ - config_name: af
39
+ data_files: data/**/af/*.parquet
40
+ - config_name: de
41
+ data_files: data/**/de/*.parquet
42
+ - config_name: default
43
+ data_files: data/**/*.parquet
44
+ - config_name: en
45
+ data_files: data/**/en/*.parquet
46
+ - config_name: es
47
+ data_files: data/**/es/*.parquet
48
+ - config_name: fr
49
+ data_files: data/**/fr/*.parquet
50
+ - config_name: fy
51
+ data_files: data/**/fy/*.parquet
52
+ - config_name: it
53
+ data_files: data/**/it/*.parquet
54
+ - config_name: nl
55
+ data_files: data/**/nl/*.parquet
56
+ dataset_info:
57
+ config_name: CC-MAIN-2024-51-af
58
+ features:
59
+ - name: text
60
+ dtype: string
61
+ - name: id
62
+ dtype: string
63
+ - name: dump
64
+ dtype: string
65
+ - name: url
66
+ dtype: string
67
+ - name: date
68
+ dtype: string
69
+ - name: file_path
70
+ dtype: string
71
+ - name: license_abbr
72
+ dtype: string
73
+ - name: license_version
74
+ dtype: string
75
+ - name: license_location
76
+ dtype: string
77
+ - name: license_in_head
78
+ dtype: bool
79
+ - name: license_in_footer
80
+ dtype: bool
81
+ - name: potential_licenses
82
+ struct:
83
+ - name: abbr
84
+ sequence: string
85
+ - name: in_footer
86
+ sequence: bool
87
+ - name: in_head
88
+ sequence: bool
89
+ - name: location
90
+ sequence: string
91
+ - name: version
92
+ sequence: string
93
+ - name: license_parse_error
94
+ dtype: bool
95
+ - name: license_disagreement
96
+ dtype: bool
97
+ - name: language
98
+ dtype: string
99
+ - name: language_score
100
+ dtype: float64
101
+ - name: in_fw2
102
+ dtype: bool
103
+ splits:
104
+ - name: train
105
+ num_bytes: 75356511
106
+ num_examples: 17733
107
+ download_size: 41860751
108
+ dataset_size: 75356511
109
+ ---
110
+
111
+ > **Raw CommonCrawl crawls, annotated with potential Creative Commons license information**
112
+
113
+ **The licensing information is extracted from the web pages based on whether they link to Creative Commons licenses but false positives may occur!** While further filtering based on the location type of the license should improve the precision (e.g. by removing hyperlink (a_tag) references), false positives may still occur.
114
+
115
+ ## Usage
116
+
117
+ ```python
118
+ from datasets import load_dataset
119
+
120
+ # Everything
121
+ ds = load_dataset("BramVanroy/CommonCrawl-CreativeCommons")
122
+
123
+ # Single dump, all languages
124
+ ds = load_dataset("BramVanroy/CommonCrawl-CreativeCommons", "CC-MAIN-2024-51")
125
+
126
+ # Single language, all dumps
127
+ ds = load_dataset("BramVanroy/CommonCrawl-CreativeCommons", "nl")
128
+
129
+ # Single language, single dump
130
+ ds = load_dataset("BramVanroy/CommonCrawl-CreativeCommons", "CC-MAIN-2024-51-nl")
131
+ ```
132
+
133
+ ## Fields
134
+
135
+ In some cases, multiple licenses are found on a single page. All licenses are collected in `potential_licenses`. From these, the "best guess" is selected
136
+ based on three criteria:
137
+
138
+ 1. location_preference_order: meta_tag, json-ld, link_tag, a_tag
139
+ 2. head_preference_order: True, False
140
+ 3. footer_preference_order: True, False
141
+
142
+ Based on these criteria, the "best guessed" license is picked as the one in the `license_*` columns. Potential disagreement between multiple licenses is given in `license_disagreement`.
143
+
144
+ - text: the extracted text (unmodified)
145
+ - id: WARC-Record-ID
146
+ - dump: Common Crawl crawl
147
+ - url: original url for document
148
+ - date: crawl date
149
+ - file_path: file path on the S3 bucket
150
+ - license_abbr: the license type. Possible values: "cc-unknown" (recommended to filter this one out), "by", "by-sa", "by-nd", "by-nc", "by-nc-sa", "by-nc-nd", "zero", "certification", "mark". If multiple licenses were found (`potential_licenses`)
151
+ - license_version: the license version, e.g. "4.0"
152
+ - license_location: the location where the license was found. Possible values: "meta_tag", "json-ld", "link_tag", "a_tag"
153
+ - license_in_head: whether the license was found inside a `head` HTML element
154
+ - license_in_footer: whether the license was found inside a `footer` HTML element, or an HTML element that had `footer` in the ID or class name
155
+ - potential_licenses:
156
+ - abbr: list of all found license abbreviations
157
+ - version: list of all found license versions
158
+ - location: list of all found license locations
159
+ - in_head: list of whether licenses were found in the head
160
+ - in_footer: list of whether licenses were found in a footer
161
+ - license_parse_error: whether there was a problem when trying to extract the license, e.g. an unparseable HTML document
162
+ - license_disagreement: whether the `potential_licenses["abbr"]` disagree, i.e., different types of licenses were found. License *versions* are not included in the comparison!
163
+ - language: the language, as detected by fastText `ft176`
164
+ - language_score: the language identification confidence score
165
+ - found_in_fw2: whether this sample was found in FineWeb-2. Crawls that are more recent than FW2 (everything after 2024-18) is marked as None
166
+
167
+
168
+ ## Progress
169
+
170
+ The attempt is to at least process all five RedPyjama crawls + `CC-MAIN-2024-51`.
171
+
172
+ Done:
173
+
174
+ - CC-MAIN-2024-51
175
+
176
+ Running:
177
+
178
+ - CC-MAIN-2019-30
179
+
180
+ To do:
181
+
182
+ - CC-MAIN-2019-30
183
+ - CC-MAIN-2020-05
184
+ - CC-MAIN-2021-04
185
+ - CC-MAIN-2022-05
186
+ - CC-MAIN-2023-06
187
+ - CC-MAIN-2024-51
188
+
189
+
190
+ ## Languages
191
+
192
+ The following languages are included.
193
+
194
+ - afr_Latn
195
+ - deu_Latn
196
+ - eng_Latn
197
+ - fra_Latn
198
+ - fry_Latn
199
+ - ita_Latn
200
+ - nld_Latn
201
+ - spa_Latn
202
+
203
+
204
+ ## Recommendations
205
+
206
+ - Raw CommonCrawl data is processed in an attempt to extract licensing information. No quality filtering is done!! It is **highly** recommended to filter this data further on quality, fluency, toxicity, etc.
207
+ - Similarly, the data has **not been deduplicated**.
208
+ - The licenses include all possible Creative Commons licenses, including non-commercial ones. Take care about what kind of data you wish to use, and filter out non-commercial licenses when needed.
209
+ - The column `license_disagreement` indicates whether multiple licenses were found that have not the same abbreviation, e.g. `cc-by` and `cc-by-nc`. It is recommended to filter these out.
210
+ - The column `license_parse_error` indicates whether an error occurred when parsing the license. You probably want to filter out documents where this was the case, though this should be extremely rare.
211
+ - Unsurpisingly, the data contains a lot of Wikipedia/Wikimedia content. Depending on what you need, you may wish to filter those out. For Wikipedia specifically, you may opt to use the more thoroughly parsed (but potentially more outdated) [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) set.
212
+ - In exceptional cases, a link to creativecommons.org is found but the exact license could not be found. These are under `license_abbr="cc-unknown"` which you may wish to filter out.
213
+
214
+ ## Acknowledgments
215
+
216
+ - [TNO](https://www.tno.nl/nl/), who funded the work hours to accomplish this collection. They intend to use parts of this material for the [GPT-NL project](https://gpt-nl.nl/).
217
+ - [Flemish Supercomputer Center](https://www.vscentrum.be/) for part of the compute under grant 2024-107
218
+ - Guilherme Penedo ([@guipenedo](https://huggingface.co/guipenedo)) and the rest of the [FineWeb](https://huggingface.co/datasets/HuggingFaceFW/fineweb) and [datatrove](https://github.com/huggingface/datatrove) team for the help and insights
219
+ - ML6 and specifically Robin Van Craenenbroek for their [Fondant Creative Commons](https://github.com/ml6team/fondant-usecase-filter-creative-commons/tree/add-fondant-usecase-cc-image-extraction) filter for image datasets. While my approach is different, their code did serve as inspiration.