Datasets:
Tasks:
Summarization
Modalities:
Text
Formats:
parquet
Sub-tasks:
news-articles-summarization
Languages:
English
Size:
100K - 1M
License:
Commit
•
898abee
1
Parent(s):
3adf624
Delete loading script
Browse files- cnn_dailymail.py +0 -250
cnn_dailymail.py
DELETED
@@ -1,250 +0,0 @@
|
|
1 |
-
# coding=utf-8
|
2 |
-
# Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
|
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 |
-
|
16 |
-
# Lint as: python3
|
17 |
-
"""CNN/DailyMail Summarization dataset, non-anonymized version."""
|
18 |
-
|
19 |
-
import hashlib
|
20 |
-
import os
|
21 |
-
|
22 |
-
import datasets
|
23 |
-
|
24 |
-
|
25 |
-
logger = datasets.logging.get_logger(__name__)
|
26 |
-
|
27 |
-
|
28 |
-
_HOMEPAGE = "https://github.com/abisee/cnn-dailymail"
|
29 |
-
|
30 |
-
_DESCRIPTION = """\
|
31 |
-
CNN/DailyMail non-anonymized summarization dataset.
|
32 |
-
|
33 |
-
There are two features:
|
34 |
-
- article: text of news article, used as the document to be summarized
|
35 |
-
- highlights: joined text of highlights with <s> and </s> around each
|
36 |
-
highlight, which is the target summary
|
37 |
-
"""
|
38 |
-
|
39 |
-
# The second citation introduces the source data, while the first
|
40 |
-
# introduces the specific form (non-anonymized) we use here.
|
41 |
-
_CITATION = """\
|
42 |
-
@article{DBLP:journals/corr/SeeLM17,
|
43 |
-
author = {Abigail See and
|
44 |
-
Peter J. Liu and
|
45 |
-
Christopher D. Manning},
|
46 |
-
title = {Get To The Point: Summarization with Pointer-Generator Networks},
|
47 |
-
journal = {CoRR},
|
48 |
-
volume = {abs/1704.04368},
|
49 |
-
year = {2017},
|
50 |
-
url = {http://arxiv.org/abs/1704.04368},
|
51 |
-
archivePrefix = {arXiv},
|
52 |
-
eprint = {1704.04368},
|
53 |
-
timestamp = {Mon, 13 Aug 2018 16:46:08 +0200},
|
54 |
-
biburl = {https://dblp.org/rec/bib/journals/corr/SeeLM17},
|
55 |
-
bibsource = {dblp computer science bibliography, https://dblp.org}
|
56 |
-
}
|
57 |
-
|
58 |
-
@inproceedings{hermann2015teaching,
|
59 |
-
title={Teaching machines to read and comprehend},
|
60 |
-
author={Hermann, Karl Moritz and Kocisky, Tomas and Grefenstette, Edward and Espeholt, Lasse and Kay, Will and Suleyman, Mustafa and Blunsom, Phil},
|
61 |
-
booktitle={Advances in neural information processing systems},
|
62 |
-
pages={1693--1701},
|
63 |
-
year={2015}
|
64 |
-
}
|
65 |
-
"""
|
66 |
-
|
67 |
-
_DL_URLS = {
|
68 |
-
"cnn_stories": "https://huggingface.co/datasets/cnn_dailymail/resolve/11343c3752184397d56efc19a8a7cceb68089318/data/cnn_stories.tgz",
|
69 |
-
"dm_stories": "https://huggingface.co/datasets/cnn_dailymail/resolve/11343c3752184397d56efc19a8a7cceb68089318/data/dailymail_stories.tgz",
|
70 |
-
"train": "https://raw.githubusercontent.com/abisee/cnn-dailymail/master/url_lists/all_train.txt",
|
71 |
-
"validation": "https://raw.githubusercontent.com/abisee/cnn-dailymail/master/url_lists/all_val.txt",
|
72 |
-
"test": "https://raw.githubusercontent.com/abisee/cnn-dailymail/master/url_lists/all_test.txt",
|
73 |
-
}
|
74 |
-
|
75 |
-
_HIGHLIGHTS = "highlights"
|
76 |
-
_ARTICLE = "article"
|
77 |
-
|
78 |
-
_SUPPORTED_VERSIONS = [
|
79 |
-
# Using cased version.
|
80 |
-
datasets.Version("3.0.0", "Using cased version."),
|
81 |
-
# Same data as 0.0.2
|
82 |
-
datasets.Version("1.0.0", ""),
|
83 |
-
# Having the model predict newline separators makes it easier to evaluate
|
84 |
-
# using summary-level ROUGE.
|
85 |
-
datasets.Version("2.0.0", "Separate target sentences with newline."),
|
86 |
-
]
|
87 |
-
|
88 |
-
|
89 |
-
_DEFAULT_VERSION = datasets.Version("3.0.0", "Using cased version.")
|
90 |
-
|
91 |
-
|
92 |
-
class CnnDailymailConfig(datasets.BuilderConfig):
|
93 |
-
"""BuilderConfig for CnnDailymail."""
|
94 |
-
|
95 |
-
def __init__(self, **kwargs):
|
96 |
-
"""BuilderConfig for CnnDailymail.
|
97 |
-
|
98 |
-
Args:
|
99 |
-
|
100 |
-
**kwargs: keyword arguments forwarded to super.
|
101 |
-
"""
|
102 |
-
super(CnnDailymailConfig, self).__init__(**kwargs)
|
103 |
-
|
104 |
-
|
105 |
-
def _get_url_hashes(path):
|
106 |
-
"""Get hashes of urls in file."""
|
107 |
-
urls = _read_text_file_path(path)
|
108 |
-
|
109 |
-
def url_hash(u):
|
110 |
-
h = hashlib.sha1()
|
111 |
-
try:
|
112 |
-
u = u.encode("utf-8")
|
113 |
-
except UnicodeDecodeError:
|
114 |
-
logger.error("Cannot hash url: %s", u)
|
115 |
-
h.update(u)
|
116 |
-
return h.hexdigest()
|
117 |
-
|
118 |
-
return {url_hash(u) for u in urls}
|
119 |
-
|
120 |
-
|
121 |
-
def _get_hash_from_path(p):
|
122 |
-
"""Extract hash from path."""
|
123 |
-
return os.path.splitext(os.path.basename(p))[0]
|
124 |
-
|
125 |
-
|
126 |
-
DM_SINGLE_CLOSE_QUOTE = "\u2019" # unicode
|
127 |
-
DM_DOUBLE_CLOSE_QUOTE = "\u201d"
|
128 |
-
# acceptable ways to end a sentence
|
129 |
-
END_TOKENS = [".", "!", "?", "...", "'", "`", '"', DM_SINGLE_CLOSE_QUOTE, DM_DOUBLE_CLOSE_QUOTE, ")"]
|
130 |
-
|
131 |
-
|
132 |
-
def _read_text_file_path(path):
|
133 |
-
with open(path, "r", encoding="utf-8") as f:
|
134 |
-
lines = [line.strip() for line in f]
|
135 |
-
return lines
|
136 |
-
|
137 |
-
|
138 |
-
def _read_text_file(file):
|
139 |
-
return [line.decode("utf-8").strip() for line in file]
|
140 |
-
|
141 |
-
|
142 |
-
def _get_art_abs(story_file, tfds_version):
|
143 |
-
"""Get abstract (highlights) and article from a story file path."""
|
144 |
-
# Based on https://github.com/abisee/cnn-dailymail/blob/master/
|
145 |
-
# make_datafiles.py
|
146 |
-
|
147 |
-
lines = _read_text_file(story_file)
|
148 |
-
|
149 |
-
# The github code lowercase the text and we removed it in 3.0.0.
|
150 |
-
|
151 |
-
# Put periods on the ends of lines that are missing them
|
152 |
-
# (this is a problem in the dataset because many image captions don't end in
|
153 |
-
# periods; consequently they end up in the body of the article as run-on
|
154 |
-
# sentences)
|
155 |
-
def fix_missing_period(line):
|
156 |
-
"""Adds a period to a line that is missing a period."""
|
157 |
-
if "@highlight" in line:
|
158 |
-
return line
|
159 |
-
if not line:
|
160 |
-
return line
|
161 |
-
if line[-1] in END_TOKENS:
|
162 |
-
return line
|
163 |
-
return line + " ."
|
164 |
-
|
165 |
-
lines = [fix_missing_period(line) for line in lines]
|
166 |
-
|
167 |
-
# Separate out article and abstract sentences
|
168 |
-
article_lines = []
|
169 |
-
highlights = []
|
170 |
-
next_is_highlight = False
|
171 |
-
for line in lines:
|
172 |
-
if not line:
|
173 |
-
continue # empty line
|
174 |
-
elif line.startswith("@highlight"):
|
175 |
-
next_is_highlight = True
|
176 |
-
elif next_is_highlight:
|
177 |
-
highlights.append(line)
|
178 |
-
else:
|
179 |
-
article_lines.append(line)
|
180 |
-
|
181 |
-
# Make article into a single string
|
182 |
-
article = " ".join(article_lines)
|
183 |
-
|
184 |
-
if tfds_version >= "2.0.0":
|
185 |
-
abstract = "\n".join(highlights)
|
186 |
-
else:
|
187 |
-
abstract = " ".join(highlights)
|
188 |
-
|
189 |
-
return article, abstract
|
190 |
-
|
191 |
-
|
192 |
-
class CnnDailymail(datasets.GeneratorBasedBuilder):
|
193 |
-
"""CNN/DailyMail non-anonymized summarization dataset."""
|
194 |
-
|
195 |
-
BUILDER_CONFIGS = [
|
196 |
-
CnnDailymailConfig(name=str(version), description="Plain text", version=version)
|
197 |
-
for version in _SUPPORTED_VERSIONS
|
198 |
-
]
|
199 |
-
|
200 |
-
def _info(self):
|
201 |
-
return datasets.DatasetInfo(
|
202 |
-
description=_DESCRIPTION,
|
203 |
-
features=datasets.Features(
|
204 |
-
{
|
205 |
-
_ARTICLE: datasets.Value("string"),
|
206 |
-
_HIGHLIGHTS: datasets.Value("string"),
|
207 |
-
"id": datasets.Value("string"),
|
208 |
-
}
|
209 |
-
),
|
210 |
-
supervised_keys=None,
|
211 |
-
homepage=_HOMEPAGE,
|
212 |
-
citation=_CITATION,
|
213 |
-
)
|
214 |
-
|
215 |
-
def _vocab_text_gen(self, paths):
|
216 |
-
for _, ex in self._generate_examples(paths):
|
217 |
-
yield " ".join([ex[_ARTICLE], ex[_HIGHLIGHTS]])
|
218 |
-
|
219 |
-
def _split_generators(self, dl_manager):
|
220 |
-
dl_paths = dl_manager.download(_DL_URLS)
|
221 |
-
return [
|
222 |
-
datasets.SplitGenerator(
|
223 |
-
name=split,
|
224 |
-
gen_kwargs={
|
225 |
-
"urls_file": dl_paths[split],
|
226 |
-
"files_per_archive": [
|
227 |
-
dl_manager.iter_archive(dl_paths["cnn_stories"]),
|
228 |
-
dl_manager.iter_archive(dl_paths["dm_stories"]),
|
229 |
-
],
|
230 |
-
},
|
231 |
-
)
|
232 |
-
for split in [datasets.Split.TRAIN, datasets.Split.VALIDATION, datasets.Split.TEST]
|
233 |
-
]
|
234 |
-
|
235 |
-
def _generate_examples(self, urls_file, files_per_archive):
|
236 |
-
urls = _get_url_hashes(urls_file)
|
237 |
-
idx = 0
|
238 |
-
for files in files_per_archive:
|
239 |
-
for path, file in files:
|
240 |
-
hash_from_path = _get_hash_from_path(path)
|
241 |
-
if hash_from_path in urls:
|
242 |
-
article, highlights = _get_art_abs(file, self.config.version)
|
243 |
-
if not article or not highlights:
|
244 |
-
continue
|
245 |
-
yield idx, {
|
246 |
-
_ARTICLE: article,
|
247 |
-
_HIGHLIGHTS: highlights,
|
248 |
-
"id": hash_from_path,
|
249 |
-
}
|
250 |
-
idx += 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|