first commit

README.md

@@ -0,0 +1,85 @@
+---
+languages:
+- en
+multilinguality:
+- monolingual
+size_categories:
+- 1K<n<10K
+task_categories:
+- conditional-text-generation
+task_ids:
+- summarization
+---
+
+# WCEP10 dataset for summarization
+
+Summarization dataset copied from [PRIMERA](https://github.com/allenai/PRIMER)
+This dataset is compatible with the [`run_summarization.py`](https://github.com/huggingface/transformers/tree/master/examples/pytorch/summarization) script from Transformers if you add this line to the `summarization_name_mapping` variable:
+```python
+"ccdv/WCEP-10": ("document", "summary")
+```
+
+# Configs
+3 possibles configs:
+- `roberta` will concatenate documents with "</s>" (default)
+- `newline` will concatenate documents with "\n"
+- `bert` will concatenate documents with "[SEP]"
+- `list` will return the list of documents instead of a string
+
+### Data Fields
+
+- `id`: paper id
+- `document`: a string/list containing the body of a set of documents
+- `summary`: a string containing the abstract of the set
+
+### Data Splits
+
+This dataset has 3 splits: _train_, _validation_, and _test_. \
+Token counts are white space based.
+
+| Dataset Split | Number of Instances | 
+| ------------- | --------------------|
+| Train         | 8158                |
+| Validation    | 1020                |
+| Test          | 1022                |
+
+
+# Cite original article
+```
+@article{DBLP:journals/corr/abs-2005-10070,
+    author    = {Demian Gholipour Ghalandari and
+                Chris Hokamp and
+                Nghia The Pham and
+                John Glover and
+                Georgiana Ifrim},
+    title     = {A Large-Scale Multi-Document Summarization Dataset from the Wikipedia
+                Current Events Portal},
+    journal   = {CoRR},
+    volume    = {abs/2005.10070},
+    year      = {2020},
+    url       = {https://arxiv.org/abs/2005.10070},
+    eprinttype = {arXiv},
+    eprint    = {2005.10070},
+    timestamp = {Fri, 22 May 2020 16:21:28 +0200},
+    biburl    = {https://dblp.org/rec/journals/corr/abs-2005-10070.bib},
+    bibsource = {dblp computer science bibliography, https://dblp.org}
+    }
+
+@article{DBLP:journals/corr/abs-2110-08499,
+    author    = {Wen Xiao and
+                Iz Beltagy and
+                Giuseppe Carenini and
+                Arman Cohan},
+    title     = {{PRIMER:} Pyramid-based Masked Sentence Pre-training for Multi-document
+                Summarization},
+    journal   = {CoRR},
+    volume    = {abs/2110.08499},
+    year      = {2021},
+    url       = {https://arxiv.org/abs/2110.08499},
+    eprinttype = {arXiv},
+    eprint    = {2110.08499},
+    timestamp = {Fri, 22 Oct 2021 13:33:09 +0200},
+    biburl    = {https://dblp.org/rec/journals/corr/abs-2110-08499.bib},
+    bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+```

WCEP-10.py

@@ -0,0 +1,158 @@
+import json
+import os
+
+import datasets
+from datasets.tasks import TextClassification
+
+_CITATION = None
+
+
+_DESCRIPTION = """
+ WCEP10 dataset for summarization.
+ From paper: "A Large-Scale Multi-Document Summarization Dataset from the Wikipedia
+                Current Events Portal" by D. Gholipour et al."
+ From paper: "PRIMER: Pyramid-based Masked Sentence Pre-training for Multi-document
+                Summarization" by W. Xiao et al."
+
+"""
+_CITATION = """\
+    @article{DBLP:journals/corr/abs-2005-10070,
+    author    = {Demian Gholipour Ghalandari and
+                Chris Hokamp and
+                Nghia The Pham and
+                John Glover and
+                Georgiana Ifrim},
+    title     = {A Large-Scale Multi-Document Summarization Dataset from the Wikipedia
+                Current Events Portal},
+    journal   = {CoRR},
+    volume    = {abs/2005.10070},
+    year      = {2020},
+    url       = {https://arxiv.org/abs/2005.10070},
+    eprinttype = {arXiv},
+    eprint    = {2005.10070},
+    timestamp = {Fri, 22 May 2020 16:21:28 +0200},
+    biburl    = {https://dblp.org/rec/journals/corr/abs-2005-10070.bib},
+    bibsource = {dblp computer science bibliography, https://dblp.org}
+    }
+
+
+    @article{DBLP:journals/corr/abs-2110-08499,
+    author    = {Wen Xiao and
+                Iz Beltagy and
+                Giuseppe Carenini and
+                Arman Cohan},
+    title     = {{PRIMER:} Pyramid-based Masked Sentence Pre-training for Multi-document
+                Summarization},
+    journal   = {CoRR},
+    volume    = {abs/2110.08499},
+    year      = {2021},
+    url       = {https://arxiv.org/abs/2110.08499},
+    eprinttype = {arXiv},
+    eprint    = {2110.08499},
+    timestamp = {Fri, 22 Oct 2021 13:33:09 +0200},
+    biburl    = {https://dblp.org/rec/journals/corr/abs-2110-08499.bib},
+    bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+"""
+_ABSTRACT = "summary"
+_ARTICLE = "document"
+
+class WCEP10SummarizationConfig(datasets.BuilderConfig):
+    """BuilderConfig for WCEP10Summarization."""
+
+    def __init__(self, **kwargs):
+        """BuilderConfig for WCEP10Summarization.
+        Args:
+          **kwargs: keyword arguments forwarded to super.
+        """
+        super(WCEP10SummarizationConfig, self).__init__(**kwargs)
+
+
+class WCEP10SummarizationDataset(datasets.GeneratorBasedBuilder):
+    """WCEP10Summarization Dataset."""
+    
+    _TRAIN_FILE = "train.zip"
+    _VAL_FILE = "val.zip"
+    _TEST_FILE = "test.zip"
+
+    BUILDER_CONFIGS = [
+        WCEP10SummarizationConfig(
+            name="newline",
+            version=datasets.Version("1.0.0"),
+            description="WCEP10 dataset for summarization, concat sections",
+        ),
+        WCEP10SummarizationConfig(
+            name="roberta",
+            version=datasets.Version("1.0.0"),
+            description="WCEP10 dataset for summarization, document",
+        ),
+        WCEP10SummarizationConfig(
+            name="bert",
+            version=datasets.Version("1.0.0"),
+            description="WCEP10 dataset for summarization, document",
+        ),
+        WCEP10SummarizationConfig(
+            name="list",
+            version=datasets.Version("1.0.0"),
+            description="WCEP10 dataset for summarization, document",
+        ),
+    ]
+
+    DEFAULT_CONFIG_NAME = "roberta"
+
+    def _info(self):
+        # Should return a datasets.DatasetInfo object
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    _ARTICLE: datasets.Sequence(datasets.Value("string")) if self.config.name == "list" else datasets.Value("string"),
+                    _ABSTRACT: datasets.Value("string"),
+                    #"id": datasets.Value("string"),
+                }
+            ),
+            supervised_keys=None,
+            homepage="https://github.com/allenai/PRIMER",
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+
+        train_path = dl_manager.download_and_extract(self._TRAIN_FILE) + "/train.txt"
+        val_path = dl_manager.download_and_extract(self._VAL_FILE) + "/val.txt"
+        test_path = dl_manager.download_and_extract(self._TEST_FILE) + "/test.txt"
+        
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN, gen_kwargs={"filepath": train_path}
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION, gen_kwargs={"filepath": val_path}
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST, gen_kwargs={"filepath": test_path}
+            ),
+        ]
+    
+    def _generate_examples(self, filepath):
+        """Generate WCEP10Summarization examples."""
+        if self.config.name == "newline":
+            join_ = "\n"
+        elif self.config.name == "roberta":
+            join_ = "</s>"
+        elif self.config.name == "bert": 
+            join_ = "[SEP]"
+
+        with open(filepath, encoding="utf-8") as f:
+            for id_, row in enumerate(f):
+                data = json.loads(row)
+
+                """
+                'summary': str,
+                'document': List[str],
+                """
+                document = data["document"]
+                if self.config.name != "list":
+                    document = join_.join(document)
+                summary = data["summary"]
+                yield id_, {"document": document, "summary": summary}

test.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c9f4e660b21b5d581eaf05c7f75ee2a8597ec9e4ac030d43b07442bf0993682
+size 6850444

train.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24e1d9989a38da9707471ec5c47d89ecd192eb34b5f678ecdc7ce2dae9411087
+size 55467667

val.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd23d8bdcb486abb41f7a14269ae472b38dcaccc108837a29c8dd9fd1cb90aec
+size 7149706