|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
"""SAMSum dataset.""" |
|
|
|
from __future__ import absolute_import, division, print_function |
|
|
|
import json |
|
|
|
import py7zr |
|
|
|
import datasets |
|
|
|
|
|
_CITATION = """ |
|
@article{gliwa2019samsum, |
|
title={SAMSum Corpus: A Human-annotated Dialogue Dataset for Abstractive Summarization}, |
|
author={Gliwa, Bogdan and Mochol, Iwona and Biesek, Maciej and Wawer, Aleksander}, |
|
journal={arXiv preprint arXiv:1911.12237}, |
|
year={2019} |
|
} |
|
""" |
|
|
|
_DESCRIPTION = """ |
|
SAMSum Corpus contains over 16k chat dialogues with manually annotated |
|
summaries. |
|
There are two features: |
|
- dialogue: text of dialogue. |
|
- summary: human written summary of the dialogue. |
|
- id: id of a example. |
|
""" |
|
|
|
_HOMEPAGE = "https://arxiv.org/abs/1911.12237v2" |
|
|
|
_LICENSE = "CC BY-NC-ND 4.0" |
|
|
|
_URLs = "https://arxiv.org/src/1911.12237v2/anc/corpus.7z" |
|
|
|
|
|
class Samsum(datasets.GeneratorBasedBuilder): |
|
"""SAMSum Corpus dataset.""" |
|
|
|
VERSION = datasets.Version("1.1.0") |
|
|
|
BUILDER_CONFIGS = [ |
|
datasets.BuilderConfig(name="samsum"), |
|
] |
|
|
|
def _info(self): |
|
features = datasets.Features( |
|
{ |
|
"id": datasets.Value("string"), |
|
"dialogue": datasets.Value("string"), |
|
"summary": datasets.Value("string"), |
|
} |
|
) |
|
return datasets.DatasetInfo( |
|
description=_DESCRIPTION, |
|
features=features, |
|
supervised_keys=None, |
|
homepage=_HOMEPAGE, |
|
license=_LICENSE, |
|
citation=_CITATION, |
|
) |
|
|
|
def _split_generators(self, dl_manager): |
|
"""Returns SplitGenerators.""" |
|
my_urls = _URLs |
|
path = dl_manager.download_and_extract(my_urls) |
|
return [ |
|
datasets.SplitGenerator( |
|
name=datasets.Split.TRAIN, |
|
|
|
gen_kwargs={ |
|
"filepath": (path, "train.json"), |
|
"split": "train", |
|
}, |
|
), |
|
datasets.SplitGenerator( |
|
name=datasets.Split.TEST, |
|
|
|
gen_kwargs={ |
|
"filepath": (path, "test.json"), |
|
"split": "test", |
|
}, |
|
), |
|
datasets.SplitGenerator( |
|
name=datasets.Split.VALIDATION, |
|
|
|
gen_kwargs={ |
|
"filepath": (path, "val.json"), |
|
"split": "val", |
|
}, |
|
), |
|
] |
|
|
|
def _generate_examples(self, filepath, split): |
|
""" Yields examples. """ |
|
|
|
path, fname = filepath |
|
|
|
with py7zr.SevenZipFile(path, "r") as z: |
|
for name, bio in z.readall().items(): |
|
if name == fname: |
|
data = json.load(bio) |
|
for example in data: |
|
yield example["id"], example |
|
|