mediasum.py

import json
import os

import datasets
from datasets.tasks import TextClassification

_CITATION = None


_DESCRIPTION = """
 MediaSum dataset for summarization.
 From paper: "MediaSum: A Large-scale Media Interview Dataset for Dialogue Summarization" by C. Zhu et al."

"""
_CITATION = """\
    @article{zhu2021mediasum,
  title={MediaSum: A Large-scale Media Interview Dataset for Dialogue Summarization},
  author={Zhu, Chenguang and Liu, Yang and Mei, Jie and Zeng, Michael},
  journal={arXiv preprint arXiv:2103.06410},
  year={2021}
}
"""
_ABSTRACT = "summary"
_ARTICLE = "document"

class MediaSumSummarizationConfig(datasets.BuilderConfig):
    """BuilderConfig for MediaSumSummarization."""

    def __init__(self, **kwargs):
        """BuilderConfig for MediaSumSummarization.
        Args:
          **kwargs: keyword arguments forwarded to super.
        """
        super(MediaSumSummarizationConfig, self).__init__(**kwargs)


class MediaSumSummarizationDataset(datasets.GeneratorBasedBuilder):
    """MediaSumSummarization Dataset."""
    
    _TRAIN_FILE = "train_data.zip"
    _VAL_FILE = "val_data.zip"
    _TEST_FILE = "test_data.zip"

    BUILDER_CONFIGS = [
        MediaSumSummarizationConfig(
            name="newline",
            version=datasets.Version("1.0.0"),
            description="MediaSum dataset for summarization, concat sections",
        ),
        MediaSumSummarizationConfig(
            name="roberta",
            version=datasets.Version("1.0.0"),
            description="MediaSum dataset for summarization, document",
        ),
        MediaSumSummarizationConfig(
            name="bert",
            version=datasets.Version("1.0.0"),
            description="MediaSum dataset for summarization, document",
        ),
        MediaSumSummarizationConfig(
            name="list",
            version=datasets.Version("1.0.0"),
            description="MediaSum dataset for summarization, document",
        ),
        MediaSumSummarizationConfig(
            name="newline_prepended",
            version=datasets.Version("1.0.0"),
            description="MediaSum dataset for summarization, concat sections",
        ),
        MediaSumSummarizationConfig(
            name="roberta_prepended",
            version=datasets.Version("1.0.0"),
            description="MediaSum dataset for summarization, document",
        ),
        MediaSumSummarizationConfig(
            name="bert_prepended",
            version=datasets.Version("1.0.0"),
            description="MediaSum dataset for summarization, document",
        ),
        MediaSumSummarizationConfig(
            name="list_prepended",
            version=datasets.Version("1.0.0"),
            description="MediaSum dataset for summarization, document",
        ),
    ]

    DEFAULT_CONFIG_NAME = "roberta_prepended"

    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/zcgzcgzcg1/MediaSum",
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):

        train_path = os.path.join(dl_manager.download_and_extract(self._TRAIN_FILE), "train_data.txt")
        val_path = os.path.join(dl_manager.download_and_extract(self._VAL_FILE), "val_data.txt")
        test_path = os.path.join(dl_manager.download_and_extract(self._TEST_FILE), "test_data.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 MediaSumSummarization examples."""
        if "newline" in self.config.name:
            join_ = "\n"
        elif "roberta" in self.config.name:
            join_ = "</s>"
        elif "bert" in self.config.name: 
            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],
                """
                
                documents = data["utt"]

                if "_prepended" in self.config.name:
                    names = data["speaker"]
                    documents = [name + ": " + document for name, document in zip(names, documents)]
                
                if self.config.name != "list":
                    documents = join_.join(documents)
                summary = data["summary"]
                yield id_, {"document": documents, "summary": summary}