first blood
Browse files- .gitignore +4 -0
- README.md +42 -0
- test.jsonl +0 -0
- wikinews-fr-100.py +130 -0
.gitignore
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
**.DS_Store
|
3 |
+
.idea
|
4 |
+
src/
|
README.md
ADDED
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Wikinews-fr-100 Benchmark Dataset for Keyphrase Generation
|
2 |
+
|
3 |
+
## About
|
4 |
+
|
5 |
+
Wikinews-fr-100 is a dataset for benchmarking keyphrase extraction and generation models.
|
6 |
+
The dataset is composed of 100 news articles in French collected from [wikinews](https://fr.wikinews.org/wiki/Accueil).
|
7 |
+
Keyphrases were annotated by readers (students in computer science) in an uncontrolled setting (that is, not limited to thesaurus entries).
|
8 |
+
Details about the dataset can be found in the original paper [(Bougouin et al., 2013)][bougouin-2013].
|
9 |
+
|
10 |
+
Reference (indexer-assigned) keyphrases are also categorized under the PRMU (<u>P</u>resent-<u>R</u>eordered-<u>M</u>ixed-<u>U</u>nseen) scheme as proposed in [(Boudin and Gallina, 2021)][boudin-2021].
|
11 |
+
|
12 |
+
Text pre-processing (tokenization) is carried out using `spacy` (`fr_core_news_sm` model) with a special rule to avoid splitting words with hyphens (e.g. graph-based is kept as one token).
|
13 |
+
Stemming (Snowball stemmer implementation for french provided in `nltk`) is applied before reference keyphrases are matched against the source text.
|
14 |
+
Details about the process can be found in `prmu.py`.
|
15 |
+
|
16 |
+
## Content and statistics
|
17 |
+
|
18 |
+
The dataset is divided into the following three splits:
|
19 |
+
|
20 |
+
| Split | # documents | #words | # keyphrases | % Present | % Reordered | % Mixed | % Unseen |
|
21 |
+
| :--------- | ----------: | -----: | -----------: | --------: | ----------: | ------: | -------: |
|
22 |
+
| Test | 100 | | | | | | |
|
23 |
+
|
24 |
+
The following data fields are available :
|
25 |
+
|
26 |
+
- **id**: unique identifier of the document.
|
27 |
+
- **title**: title of the document.
|
28 |
+
- **abstract**: abstract of the document.
|
29 |
+
- **keyphrases**: list of reference keyphrases.
|
30 |
+
- **prmu**: list of <u>P</u>resent-<u>R</u>eordered-<u>M</u>ixed-<u>U</u>nseen categories for reference keyphrases.
|
31 |
+
|
32 |
+
## References
|
33 |
+
|
34 |
+
- (Bougouin et al., 2013) Adrien Bougouin, Florian Boudin, and Béatrice Daille. 2013.
|
35 |
+
[TopicRank: Graph-Based Topic Ranking for Keyphrase Extraction](https://aclanthology.org/I13-1062/).
|
36 |
+
In Proceedings of the Sixth International Joint Conference on Natural Language Processing, pages 543–551, Nagoya, Japan. Asian Federation of Natural Language Processing.
|
37 |
+
- (Boudin and Gallina, 2021) Florian Boudin and Ygor Gallina. 2021.
|
38 |
+
[Redefining Absent Keyphrases and their Effect on Retrieval Effectiveness](https://aclanthology.org/2021.naacl-main.330/).
|
39 |
+
In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4185–4193, Online. Association for Computational Linguistics.
|
40 |
+
|
41 |
+
[bougouin-2013]: https://aclanthology.org/I13-1062/
|
42 |
+
[boudin-2021]: https://aclanthology.org/2021.naacl-main.330/
|
test.jsonl
ADDED
The diff for this file is too large to render.
See raw diff
|
|
wikinews-fr-100.py
ADDED
@@ -0,0 +1,130 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""Inspec benchmark dataset for keyphrase extraction an generation."""
|
2 |
+
|
3 |
+
import csv
|
4 |
+
import json
|
5 |
+
import os
|
6 |
+
import datasets
|
7 |
+
|
8 |
+
# TODO: Add BibTeX citation
|
9 |
+
# Find for instance the citation on arxiv or on the dataset repo/website
|
10 |
+
_CITATION = """\
|
11 |
+
@inproceedings{bougouin-etal-2013-topicrank,
|
12 |
+
title = "{T}opic{R}ank: Graph-Based Topic Ranking for Keyphrase Extraction",
|
13 |
+
author = "Bougouin, Adrien and
|
14 |
+
Boudin, Florian and
|
15 |
+
Daille, B{\'e}atrice",
|
16 |
+
booktitle = "Proceedings of the Sixth International Joint Conference on Natural Language Processing",
|
17 |
+
month = oct,
|
18 |
+
year = "2013",
|
19 |
+
address = "Nagoya, Japan",
|
20 |
+
publisher = "Asian Federation of Natural Language Processing",
|
21 |
+
url = "https://aclanthology.org/I13-1062",
|
22 |
+
pages = "543--551",
|
23 |
+
}
|
24 |
+
"""
|
25 |
+
|
26 |
+
# You can copy an official description
|
27 |
+
_DESCRIPTION = """\
|
28 |
+
Wikinews-fr-100 benchmark dataset for keyphrase extraction an generation.
|
29 |
+
"""
|
30 |
+
|
31 |
+
# TODO: Add a link to an official homepage for the dataset here
|
32 |
+
_HOMEPAGE = "https://aclanthology.org/I13-1062.pdf"
|
33 |
+
|
34 |
+
# TODO: Add the licence for the dataset here if you can find it
|
35 |
+
_LICENSE = "Apache 2.0 License"
|
36 |
+
|
37 |
+
# TODO: Add link to the official dataset URLs here
|
38 |
+
# The HuggingFace Datasets library doesn't host the datasets but only points to the original files.
|
39 |
+
# This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
|
40 |
+
_URLS = {
|
41 |
+
"test": "test.jsonl"
|
42 |
+
}
|
43 |
+
|
44 |
+
# TODO: Name of the dataset usually match the script name with CamelCase instead of snake_case
|
45 |
+
class Inspec(datasets.GeneratorBasedBuilder):
|
46 |
+
"""TODO: Short description of my dataset."""
|
47 |
+
|
48 |
+
VERSION = datasets.Version("1.0.0")
|
49 |
+
|
50 |
+
# This is an example of a dataset with multiple configurations.
|
51 |
+
# If you don't want/need to define several sub-sets in your dataset,
|
52 |
+
# just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.
|
53 |
+
|
54 |
+
# If you need to make complex sub-parts in the datasets with configurable options
|
55 |
+
# You can create your own builder configuration class to store attribute, inheriting from datasets.BuilderConfig
|
56 |
+
# BUILDER_CONFIG_CLASS = MyBuilderConfig
|
57 |
+
|
58 |
+
# You will be able to load one or the other configurations in the following list with
|
59 |
+
# data = datasets.load_dataset('my_dataset', 'first_domain')
|
60 |
+
# data = datasets.load_dataset('my_dataset', 'second_domain')
|
61 |
+
BUILDER_CONFIGS = [
|
62 |
+
datasets.BuilderConfig(name="raw", version=VERSION, description="This part of my dataset covers the raw data."),
|
63 |
+
]
|
64 |
+
|
65 |
+
DEFAULT_CONFIG_NAME = "raw" # It's not mandatory to have a default configuration. Just use one if it make sense.
|
66 |
+
|
67 |
+
def _info(self):
|
68 |
+
# TODO: This method specifies the datasets.DatasetInfo object which contains informations and typings for the dataset
|
69 |
+
if self.config.name == "raw": # This is the name of the configuration selected in BUILDER_CONFIGS above
|
70 |
+
features = datasets.Features(
|
71 |
+
{
|
72 |
+
"id": datasets.Value("int64"),
|
73 |
+
"title": datasets.Value("string"),
|
74 |
+
"abstract": datasets.Value("string"),
|
75 |
+
"keyphrases": datasets.features.Sequence(datasets.Value("string")),
|
76 |
+
"prmu": datasets.features.Sequence(datasets.Value("string")),
|
77 |
+
}
|
78 |
+
)
|
79 |
+
return datasets.DatasetInfo(
|
80 |
+
# This is the description that will appear on the datasets page.
|
81 |
+
description=_DESCRIPTION,
|
82 |
+
# This defines the different columns of the dataset and their types
|
83 |
+
features=features, # Here we define them above because they are different between the two configurations
|
84 |
+
# If there's a common (input, target) tuple from the features, uncomment supervised_keys line below and
|
85 |
+
# specify them. They'll be used if as_supervised=True in builder.as_dataset.
|
86 |
+
# supervised_keys=("sentence", "label"),
|
87 |
+
# Homepage of the dataset for documentation
|
88 |
+
homepage=_HOMEPAGE,
|
89 |
+
# License for the dataset if available
|
90 |
+
license=_LICENSE,
|
91 |
+
# Citation for the dataset
|
92 |
+
citation=_CITATION,
|
93 |
+
)
|
94 |
+
|
95 |
+
def _split_generators(self, dl_manager):
|
96 |
+
# TODO: This method is tasked with downloading/extracting the data and defining the splits depending on the configuration
|
97 |
+
# If several configurations are possible (listed in BUILDER_CONFIGS), the configuration selected by the user is in self.config.name
|
98 |
+
|
99 |
+
# dl_manager is a datasets.download.DownloadManager that can be used to download and extract URLS
|
100 |
+
# It can accept any type or nested list/dict and will give back the same structure with the url replaced with path to local files.
|
101 |
+
# By default the archives will be extracted and a path to a cached folder where they are extracted is returned instead of the archive
|
102 |
+
urls = _URLS
|
103 |
+
data_dir = dl_manager.download_and_extract(urls)
|
104 |
+
return [
|
105 |
+
datasets.SplitGenerator(
|
106 |
+
name=datasets.Split.TEST,
|
107 |
+
# These kwargs will be passed to _generate_examples
|
108 |
+
gen_kwargs={
|
109 |
+
"filepath": os.path.join(data_dir["test"]),
|
110 |
+
"split": "test"
|
111 |
+
},
|
112 |
+
),
|
113 |
+
]
|
114 |
+
|
115 |
+
# method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
|
116 |
+
def _generate_examples(self, filepath, split):
|
117 |
+
# TODO: This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
|
118 |
+
# The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
|
119 |
+
with open(filepath, encoding="utf-8") as f:
|
120 |
+
for key, row in enumerate(f):
|
121 |
+
data = json.loads(row)
|
122 |
+
# Yields examples as (key, example) tuples
|
123 |
+
yield key, {
|
124 |
+
"id": data["id"],
|
125 |
+
"title": data["title"],
|
126 |
+
"abstract": data["abstract"],
|
127 |
+
"keyphrases": data["keyphrases"],
|
128 |
+
"prmu": data["prmu"],
|
129 |
+
}
|
130 |
+
|