Copy from CovidFact.

README.md

@@ -0,0 +1,84 @@
+---
+annotations_creators:
+- expert-generated
+language_creators:
+- found
+language:
+- en
+license:
+- cc-by-nc-2.0
+multilinguality:
+- monolingual
+size_categories:
+- 1K<n<10K
+source_datasets:
+- original
+task_categories:
+- text-classification
+task_ids:
+- fact-checking
+pretty_name: CovidFact
+dataset_info:
+  features:
+  - name: claim_id
+    dtype: int32
+  - name: claim
+    dtype: string
+  - name: abstract_id
+    dtype: int32
+  - name: title
+    dtype: string
+  - name: abstract
+    sequence: string
+  - name: verdict
+    dtype: string
+  - name: evidence
+    sequence: int32
+  splits:
+  - name: train
+    num_bytes: 1547185
+    num_examples: 940
+  - name: test
+    num_bytes: 523542
+    num_examples: 317
+  download_size: 3610222
+  dataset_size: 2070727
+---
+
+
+# Dataset Card for "covidfact_entailment"
+
+## Table of Contents
+
+- [Dataset Description](#dataset-description)
+  - [Dataset Summary](#dataset-summary)
+- [Dataset Structure](#dataset-structure)
+  - [Data Fields](#data-fields)
+  - [Data Splits](#data-splits)
+
+## Dataset Description
+
+- **Repository:** <https://github.com/asaakyan/covidfact>
+- **Point of Contact:** [David Wadden](mailto:davidw@allenai.org)
+
+### Dataset Summary
+
+COVID-FACT is a dataset of claims about COVID-19. For this version of the dataset, we follow the preprocessing from the MultiVerS modeling paper https://github.com/dwadden/multivers, verifying claims against abstracts of scientific research articles. Entailment labels and rationales are included.
+
+## Dataset Structure
+
+### Data fields
+
+- `claim_id`: An `int32` claim identifier.
+- `claim`: A `string`.
+- `abstract_id`: An `int32` abstract identifier.
+- `title`: A `string`.
+- `abstract`: A list of `strings`, one for each sentence in the abstract.
+- `verdict`: The fact-checking verdict, a `string`.
+- `evidence`: A list of sentences from the abstract which provide evidence for the verdict.
+
+### Data Splits
+
+|      |train|validation|
+|------|----:|---------:|
+|claims| 919 |       340|

healthver_entailment.py

@@ -0,0 +1,161 @@
+"""Scientific fact-checking dataset. Verifies claims based on citation sentences
+using evidence from the cited abstracts. Formatted as a paragraph-level entailment task."""
+
+
+import datasets
+import json
+
+
+_CITATION = """\
+@article{Saakyan2021COVIDFactFE,
+  title={COVID-Fact: Fact Extraction and Verification of Real-World Claims on COVID-19 Pandemic},
+  author={Arkadiy Saakyan and Tuhin Chakrabarty and Smaranda Muresan},
+  journal={ArXiv},
+  year={2021},
+  volume={abs/2106.03794},
+  url={https://api.semanticscholar.org/CorpusID:235364036}
+}
+"""
+
+
+_DESCRIPTION = """\
+COVID-FACT is a dataset of claims about COVID-19. For this version of the dataset, we follow the preprocessing from the MultiVerS modeling paper https://github.com/dwadden/multivers, verifying claims against abstracts of scientific research articles. Entailment labels and rationales are included.
+"""
+
+_URL = "https://scifact.s3.us-west-2.amazonaws.com/longchecker/latest/data.tar.gz"
+
+
+def flatten(xss):
+    return [x for xs in xss for x in xs]
+
+
+class CovidFactEntailmentConfig(datasets.BuilderConfig):
+    """builderconfig for covidfact"""
+
+    def __init__(self, **kwargs):
+        """
+
+        Args:
+            **kwargs: keyword arguments forwarded to super.
+        """
+        super(CovidFactEntailmentConfig, self).__init__(
+            version=datasets.Version("1.0.0", ""), **kwargs
+        )
+
+
+class CovidFactEntailment(datasets.GeneratorBasedBuilder):
+    """TODO(covidfact): Short description of my dataset."""
+
+    # TODO(covidfact): Set up version.
+    VERSION = datasets.Version("0.1.0")
+
+    def _info(self):
+        # TODO(covidfact): Specifies the datasets.DatasetInfo object
+
+        features = {
+            "claim_id": datasets.Value("int32"),
+            "claim": datasets.Value("string"),
+            "abstract_id": datasets.Value("int32"),
+            "title": datasets.Value("string"),
+            "abstract": datasets.features.Sequence(datasets.Value("string")),
+            "verdict": datasets.Value("string"),
+            "evidence": datasets.features.Sequence(datasets.Value("int32")),
+        }
+
+        return datasets.DatasetInfo(
+            # This is the description that will appear on the datasets page.
+            description=_DESCRIPTION,
+            # datasets.features.FeatureConnectors
+            features=datasets.Features(
+                features
+                # These are the features of your dataset like images, labels ...
+            ),
+            # If there's a common (input, target) tuple from the features,
+            # specify them here. They'll be used if as_supervised=True in
+            # builder.as_dataset.
+            supervised_keys=None,
+            # Homepage of the dataset for documentation
+            homepage="https://scifact.apps.allenai.org/",
+            citation=_CITATION,
+        )
+
+    @staticmethod
+    def _read_tar_file(f):
+        res = []
+        for row in f:
+            this_row = json.loads(row.decode("utf-8"))
+            res.append(this_row)
+
+        return res
+
+    def _split_generators(self, dl_manager):
+        """Returns SplitGenerators."""
+        # TODO(scifact): Downloads the data and defines the splits
+        # dl_manager is a datasets.download.DownloadManager that can be used to
+        # download and extract URLs
+        archive = dl_manager.download(_URL)
+        for path, f in dl_manager.iter_archive(archive):
+            # The claims are too similar to paper titles; don't include.
+            if path == "data/covidfact/corpus_without_titles.jsonl":
+                corpus = self._read_tar_file(f)
+                corpus = {x["doc_id"]: x for x in corpus}
+            elif path == "data/covidfact/claims_train.jsonl":
+                claims_train = self._read_tar_file(f)
+            elif path == "data/covidfact/claims_test.jsonl":
+                claims_test = self._read_tar_file(f)
+
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "claims": claims_train,
+                    "corpus": corpus,
+                    "split": "train",
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "claims": claims_test,
+                    "corpus": corpus,
+                    "split": "test",
+                },
+            ),
+        ]
+
+    def _generate_examples(self, claims, corpus, split):
+        """Yields examples."""
+        # Loop over claims and put evidence together with claim.
+        id_ = -1  # Will increment to 0 on first iteration.
+        for claim in claims:
+            evidence = {int(k): v for k, v in claim["evidence"].items()}
+            for cited_doc_id in claim["doc_ids"]:
+                cited_doc = corpus[cited_doc_id]
+                abstract_sents = [sent.strip() for sent in cited_doc["abstract"]]
+
+                if cited_doc_id in evidence:
+                    this_evidence = evidence[cited_doc_id]
+                    verdict = this_evidence[0][
+                        "label"
+                    ]  # Can take first evidence since all labels are same.
+                    evidence_sents = flatten(
+                        [entry["sentences"] for entry in this_evidence]
+                    )
+                else:
+                    verdict = "NEI"
+                    evidence_sents = []
+
+                instance = {
+                    "claim_id": claim["id"],
+                    "claim": claim["claim"],
+                    "abstract_id": cited_doc_id,
+                    "title": cited_doc["title"],
+                    "abstract": abstract_sents,
+                    "verdict": verdict,
+                    "evidence": evidence_sents,
+                }
+
+                id_ += 1
+                yield id_, instance