AMIsum.py

import json
import os

import datasets


class ERRNewsConfig(datasets.BuilderConfig):
    def __init__(self, data_url, features, **kwargs):
        super().__init__(version=datasets.Version("1.0.0"), **kwargs)
        self.features = features
        self.data_url = data_url


class ERRNews(datasets.GeneratorBasedBuilder):
    features = ["transcript", "summary", "id"]
    data_url = "https://cs.taltech.ee/staff/heharm/AMIsum/"

    BUILDER_CONFIGS = [
        ERRNewsConfig(
            name="full",
            features=features,
            data_url=data_url
        )
    ]

    DEFAULT_CONFIG_NAME = "full"

    def _info(self):
        features = datasets.Features(
            {
                "transcript": datasets.Value("string"),
                "summary": datasets.Value("string"),
                "id": datasets.Value("string"),
            })

        description = """\
        AMI Summarization Dataset (AMIsum) is a meeting summarization dataset, consisting of meeting transcripts \
        and abstract summaries from the AMI Corpus: https://groups.inf.ed.ac.uk/ami/corpus/.
        """

        return datasets.DatasetInfo(
            features=features,
            description=description,
            supervised_keys=None,
            version=self.config.version,
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        train = "train.json"
        test = "test.json"
        val = "val.json"

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "file_path": dl_manager.download(self.config.data_url + train),
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "file_path": dl_manager.download(self.config.data_url + val),
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "file_path": dl_manager.download(self.config.data_url + test),
                },
            ),
        ]

    def create_dict(self, data):
        res = dict()
        for key in self.config.features:
            res[key] = data[key]
        return res

    def _generate_examples(self, file_path):
        with open(file_path) as f:
            data = json.load(f)
            for idx, transcript in enumerate(data["transcript"]):
                id_ = data["id"][idx]
                yield id_, {
                    "transcript": transcript,
                    "summary": data["summary"][idx],
                    "id": id_,
                }