Datasets:
Tasks:
Text Classification
Languages:
English
Size:
100K<n<1M
ArXiv:
Tags:
knowledge-verification
License:
Add dataset loading script
Browse files- feverous.py +126 -0
feverous.py
ADDED
@@ -0,0 +1,126 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""FEVEROUS dataset."""
|
2 |
+
|
3 |
+
import json
|
4 |
+
import textwrap
|
5 |
+
|
6 |
+
import datasets
|
7 |
+
|
8 |
+
|
9 |
+
class FeverousConfig(datasets.BuilderConfig):
|
10 |
+
"""BuilderConfig for FEVER."""
|
11 |
+
|
12 |
+
def __init__(self, homepage: str = None, citation: str = None, base_url: str = None, urls: dict = None, **kwargs):
|
13 |
+
"""BuilderConfig for FEVEROUS.
|
14 |
+
|
15 |
+
Args:
|
16 |
+
homepage (`str`): Homepage.
|
17 |
+
citation (`str`): Citation reference.
|
18 |
+
base_url (`str`): Data base URL that precedes all data URLs.
|
19 |
+
urls (`dict`): Data URLs (each URL will pe preceded by `base_url`).
|
20 |
+
**kwargs: keyword arguments forwarded to super.
|
21 |
+
"""
|
22 |
+
super().__init__(**kwargs)
|
23 |
+
self.homepage = homepage
|
24 |
+
self.citation = citation
|
25 |
+
self.base_url = base_url
|
26 |
+
self.urls = {key: f"{base_url}/{url}" for key, url in urls.items()}
|
27 |
+
|
28 |
+
|
29 |
+
class FeverOUS(datasets.GeneratorBasedBuilder):
|
30 |
+
"""FEVEROUS dataset."""
|
31 |
+
|
32 |
+
BUILDER_CONFIGS = [
|
33 |
+
FeverousConfig(
|
34 |
+
version=datasets.Version("1.0.0"),
|
35 |
+
description=textwrap.dedent(
|
36 |
+
"FEVEROUS:\n"
|
37 |
+
"FEVEROUS (Fact Extraction and VERification Over Unstructured and Structured information) is a fact "
|
38 |
+
"verification dataset which consists of 87,026 verified claims. Each claim is annotated with evidence "
|
39 |
+
"in the form of sentences and/or cells from tables in Wikipedia, as well as a label indicating whether "
|
40 |
+
"this evidence supports, refutes, or does not provide enough information to reach a verdict. The "
|
41 |
+
"dataset also contains annotation metadata such as annotator actions (query keywords, clicks on page, "
|
42 |
+
"time signatures), and the type of challenge each claim poses."
|
43 |
+
),
|
44 |
+
homepage="https://fever.ai/dataset/feverous.html",
|
45 |
+
citation=textwrap.dedent(
|
46 |
+
"""\
|
47 |
+
@inproceedings{Aly21Feverous,
|
48 |
+
author = {Aly, Rami and Guo, Zhijiang and Schlichtkrull, Michael Sejr and Thorne, James and Vlachos, Andreas and Christodoulopoulos, Christos and Cocarascu, Oana and Mittal, Arpit},
|
49 |
+
title = {{FEVEROUS}: Fact Extraction and {VERification} Over Unstructured and Structured information},
|
50 |
+
eprint={2106.05707},
|
51 |
+
archivePrefix={arXiv},
|
52 |
+
primaryClass={cs.CL},
|
53 |
+
year = {2021}
|
54 |
+
}"""
|
55 |
+
),
|
56 |
+
base_url="https://fever.ai/download/feverous",
|
57 |
+
urls={
|
58 |
+
datasets.Split.TRAIN: "feverous_train_challenges.jsonl",
|
59 |
+
datasets.Split.VALIDATION: "feverous_dev_challenges.jsonl",
|
60 |
+
datasets.Split.TEST: "feverous_test_unlabeled.jsonl",
|
61 |
+
},
|
62 |
+
),
|
63 |
+
]
|
64 |
+
|
65 |
+
def _info(self):
|
66 |
+
features = {
|
67 |
+
"id": datasets.Value("int32"),
|
68 |
+
"label": datasets.ClassLabel(names=["SUPPORTS", "REFUTES", "NOT ENOUGH INFO"]),
|
69 |
+
"claim": datasets.Value("string"),
|
70 |
+
"evidence": [
|
71 |
+
{
|
72 |
+
"content": [datasets.Value("string")],
|
73 |
+
"context": [[datasets.Value("string")]],
|
74 |
+
}
|
75 |
+
],
|
76 |
+
"annotator_operations": [
|
77 |
+
{
|
78 |
+
"operation": datasets.Value("string"),
|
79 |
+
"value": datasets.Value("string"),
|
80 |
+
"time": datasets.Value("float"),
|
81 |
+
}
|
82 |
+
],
|
83 |
+
"expected_challenge": datasets.Value("string"),
|
84 |
+
"challenge": datasets.Value("string"),
|
85 |
+
}
|
86 |
+
return datasets.DatasetInfo(
|
87 |
+
description=self.config.description,
|
88 |
+
features=datasets.Features(features),
|
89 |
+
homepage=self.config.homepage,
|
90 |
+
citation=self.config.citation,
|
91 |
+
)
|
92 |
+
|
93 |
+
def _split_generators(self, dl_manager):
|
94 |
+
dl_paths = dl_manager.download_and_extract(self.config.urls)
|
95 |
+
return [
|
96 |
+
datasets.SplitGenerator(
|
97 |
+
name=split,
|
98 |
+
gen_kwargs={
|
99 |
+
"filepath": dl_paths[split],
|
100 |
+
},
|
101 |
+
)
|
102 |
+
for split in dl_paths.keys()
|
103 |
+
]
|
104 |
+
|
105 |
+
def _generate_examples(self, filepath):
|
106 |
+
with open(filepath, encoding="utf-8") as f:
|
107 |
+
for id_, row in enumerate(f):
|
108 |
+
data = json.loads(row)
|
109 |
+
# First item in "train" has all values equal to empty strings
|
110 |
+
if [value for value in data.values() if value]:
|
111 |
+
evidence = data.get("evidence", [])
|
112 |
+
if evidence:
|
113 |
+
for evidence_set in evidence:
|
114 |
+
# Transform "context" from dict to list (analogue to "content")
|
115 |
+
evidence_set["context"] = [
|
116 |
+
evidence_set["context"][element_id] for element_id in evidence_set["content"]
|
117 |
+
]
|
118 |
+
yield id_, {
|
119 |
+
"id": data.get("id"),
|
120 |
+
"label": data.get("label", -1),
|
121 |
+
"claim": data.get("claim", ""),
|
122 |
+
"evidence": evidence,
|
123 |
+
"annotator_operations": data.get("annotator_operations", []),
|
124 |
+
"expected_challenge": data.get("expected_challenge", ""),
|
125 |
+
"challenge": data.get("challenge", ""),
|
126 |
+
}
|