Upload scifact.py
Browse filesAdding a scifact data loader script.
- scifact.py +161 -0
scifact.py
ADDED
@@ -0,0 +1,161 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import datasets
|
3 |
+
|
4 |
+
_DESCRIPTION = """
|
5 |
+
SciFact
|
6 |
+
|
7 |
+
A dataset of expert-written scientific claims paired with evidence-containing
|
8 |
+
abstracts and annotated with labels and rationales.
|
9 |
+
"""
|
10 |
+
|
11 |
+
_CITATION = """
|
12 |
+
@InProceedings{Wadden2020FactOF,
|
13 |
+
author = {David Wadden, Shanchuan Lin, Kyle Lo, Lucy Lu Wang,
|
14 |
+
Madeleine van Zuylen, Arman Cohan, Hannaneh Hajishirzi},
|
15 |
+
title = {Fact or Fiction: Verifying Scientific Claims},
|
16 |
+
booktitle = {EMNLP},
|
17 |
+
year = 2020,
|
18 |
+
}
|
19 |
+
"""
|
20 |
+
|
21 |
+
_DOWNLOAD_URL = "https://testerstories.com/files/ai_learn/data.tar.gz"
|
22 |
+
|
23 |
+
|
24 |
+
class ScifactConfig(datasets.BuilderConfig):
|
25 |
+
def __init__(self, **kwargs):
|
26 |
+
super(ScifactConfig, self).__init__(
|
27 |
+
version=datasets.Version("1.0.0", ""), **kwargs
|
28 |
+
)
|
29 |
+
|
30 |
+
|
31 |
+
class Scifact(datasets.GeneratorBasedBuilder):
|
32 |
+
VERSION = datasets.Version("0.1.0")
|
33 |
+
|
34 |
+
BUILDER_CONFIGS = [
|
35 |
+
ScifactConfig(name="corpus", description="The corpus of evidence documents"),
|
36 |
+
ScifactConfig(
|
37 |
+
name="claims", description="The claims are split into train, test, dev"
|
38 |
+
),
|
39 |
+
]
|
40 |
+
|
41 |
+
def _info(self):
|
42 |
+
if self.config.name == "corpus":
|
43 |
+
features = {
|
44 |
+
"doc_id": datasets.Value("int32"),
|
45 |
+
"title": datasets.Value("string"),
|
46 |
+
"abstract": datasets.features.Sequence(datasets.Value("string")),
|
47 |
+
"structured": datasets.Value("bool"),
|
48 |
+
}
|
49 |
+
else:
|
50 |
+
features = {
|
51 |
+
"id": datasets.Value("int32"),
|
52 |
+
"claim": datasets.Value("string"),
|
53 |
+
"evidence_doc_id": datasets.Value("string"),
|
54 |
+
"evidence_label": datasets.Value("string"),
|
55 |
+
"evidence_sentences": datasets.features.Sequence(
|
56 |
+
datasets.Value("int32")
|
57 |
+
),
|
58 |
+
"cited_doc_ids": datasets.features.Sequence(datasets.Value("int32")),
|
59 |
+
}
|
60 |
+
|
61 |
+
return datasets.DatasetInfo(
|
62 |
+
description=_DESCRIPTION,
|
63 |
+
features=datasets.Features(features),
|
64 |
+
supervised_keys=None,
|
65 |
+
homepage="https://scifact.apps.allenai.org/",
|
66 |
+
citation=_CITATION,
|
67 |
+
)
|
68 |
+
|
69 |
+
def _split_generators(self, dl_manager):
|
70 |
+
archive = dl_manager.download(_DOWNLOAD_URL)
|
71 |
+
|
72 |
+
if self.config.name == "corpus":
|
73 |
+
return [
|
74 |
+
datasets.SplitGenerator(
|
75 |
+
name=datasets.Split.TRAIN,
|
76 |
+
gen_kwargs={
|
77 |
+
"filepath": "data/corpus.jsonl",
|
78 |
+
"split": "train",
|
79 |
+
"files": dl_manager.iter_archive(archive),
|
80 |
+
},
|
81 |
+
),
|
82 |
+
]
|
83 |
+
else:
|
84 |
+
return [
|
85 |
+
datasets.SplitGenerator(
|
86 |
+
name=datasets.Split.TRAIN,
|
87 |
+
gen_kwargs={
|
88 |
+
"filepath": "data/claims_train.jsonl",
|
89 |
+
"split": "train",
|
90 |
+
"files": dl_manager.iter_archive(archive),
|
91 |
+
},
|
92 |
+
),
|
93 |
+
datasets.SplitGenerator(
|
94 |
+
name=datasets.Split.TEST,
|
95 |
+
gen_kwargs={
|
96 |
+
"filepath": "data/claims_test.jsonl",
|
97 |
+
"split": "test",
|
98 |
+
"files": dl_manager.iter_archive(archive),
|
99 |
+
},
|
100 |
+
),
|
101 |
+
datasets.SplitGenerator(
|
102 |
+
name=datasets.Split.VALIDATION,
|
103 |
+
gen_kwargs={
|
104 |
+
"filepath": "data/claims_dev.jsonl",
|
105 |
+
"split": "dev",
|
106 |
+
"files": dl_manager.iter_archive(archive),
|
107 |
+
},
|
108 |
+
),
|
109 |
+
]
|
110 |
+
|
111 |
+
def _generate_examples(self, filepath, split, files):
|
112 |
+
for path, f in files:
|
113 |
+
if path == filepath:
|
114 |
+
for id_, row in enumerate(f):
|
115 |
+
data = json.loads(row.decode("utf-8"))
|
116 |
+
|
117 |
+
if self.config.name == "corpus":
|
118 |
+
yield id_, {
|
119 |
+
"doc_id": int(data["doc_id"]),
|
120 |
+
"title": data["title"],
|
121 |
+
"abstract": data["abstract"],
|
122 |
+
"structured": data["structured"],
|
123 |
+
}
|
124 |
+
else:
|
125 |
+
if split == "test":
|
126 |
+
yield id_, {
|
127 |
+
"id": data["id"],
|
128 |
+
"claim": data["claim"],
|
129 |
+
"evidence_doc_id": "",
|
130 |
+
"evidence_label": "",
|
131 |
+
"evidence_sentences": [],
|
132 |
+
"cited_doc_ids": [],
|
133 |
+
}
|
134 |
+
else:
|
135 |
+
evidences = data["evidence"]
|
136 |
+
|
137 |
+
if evidences:
|
138 |
+
for id1, doc_id in enumerate(evidences):
|
139 |
+
for id2, evidence in enumerate(evidences[doc_id]):
|
140 |
+
yield str(id_) + "_" + str(id1) + "_" + str(
|
141 |
+
id2
|
142 |
+
), {
|
143 |
+
"id": data["id"],
|
144 |
+
"claim": data["claim"],
|
145 |
+
"evidence_doc_id": doc_id,
|
146 |
+
"evidence_label": evidence["label"],
|
147 |
+
"evidence_sentences": evidence["sentences"],
|
148 |
+
"cited_doc_ids": data.get(
|
149 |
+
"cited_doc_ids", []
|
150 |
+
),
|
151 |
+
}
|
152 |
+
else:
|
153 |
+
yield id_, {
|
154 |
+
"id": data["id"],
|
155 |
+
"claim": data["claim"],
|
156 |
+
"evidence_doc_id": "",
|
157 |
+
"evidence_label": "",
|
158 |
+
"evidence_sentences": [],
|
159 |
+
"cited_doc_ids": data.get("cited_doc_ids", []),
|
160 |
+
}
|
161 |
+
break
|