TempoSum.py

"""The TempoSum benchmark."""

import json
import os
import datasets

from contextlib import ExitStack

_CITATION = """
@misc{cheang2023temposum,
Author = {Chi Seng Cheang and Hou Pong Chan and Derek F. Wong and Xuebo Liu and Zhaocong Li and Yanming Sun and Shudong Liu and Lidia S. Chao},
Title = {TempoSum: Evaluating the Temporal Generalization of Abstractive Summarization},
Year = {2023},
}
"""

_DESCRIPTION = """TempoSum: Evaluating the Temporal Generalization of Abstractive Summarization"""

_URL = "https://huggingface.co/datasets/chiseng-cheang/TempoSum/resolve/main/data/"

_DOCUMENT = "document"
_SUMMARY = "summary"
_TITLE = "title"

_DATASET_CONFIGS = {
    "BBC_in-distribution": {
        "urls": {
            datasets.Split.TEST: os.path.join(_URL, "bbc_in_distribution.tar.gz"),
        },
        "available_features": [_DOCUMENT, _SUMMARY],
    },
    "BBC_future": {
        "urls": {
            datasets.Split.TEST: os.path.join(_URL, "bbc_future.tar.gz"),
        },
        "available_features": [_DOCUMENT, _SUMMARY],
    },
    "CNN_in-distribution": {
        "urls": {
            datasets.Split.TEST: os.path.join(_URL, "cnn_in_distribution.tar.gz"),
        },
        "available_features": [_DOCUMENT, _TITLE],
    },
    "CNN_future": {
        "urls": {
            datasets.Split.TEST: os.path.join(_URL, "cnn_future.tar.gz"),
        },
        "available_features": [_DOCUMENT, _TITLE],
    },
}


class TempoSumConfig(datasets.BuilderConfig):
    """BuilderConfig for TempoSum."""
    def __init__(self, urls, available_features, **kwargs):
        super(TempoSumConfig, self).__init__(version=datasets.Version("1.0.0"), **kwargs)
        self.features = datasets.Features({
            feature: datasets.Value("string") for feature in available_features
            # _DOCUMENT: datasets.Value("string"),
            # _SUMMARY: datasets.Value("string"),
        })
        self.urls = urls
        self.available_features = available_features

class TempoSum(datasets.GeneratorBasedBuilder):
    """The TempoSum benchmark."""
    BUILDER_CONFIGS = []

    for datasplit_name, datasplit_config in _DATASET_CONFIGS.items():
        BUILDER_CONFIGS.append(
            TempoSumConfig(
                name=datasplit_name,
                urls=datasplit_config['urls'],
                available_features=datasplit_config['available_features'],
            )
        )

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            homepage="https://github.com/AndyCheang/TempoSum",
        )

    def _split_generators(self, dl_manager):
        dl_dirs = dl_manager.download_and_extract(self.config.urls)
        splits = []
        for split in dl_dirs:
            splits.append(
                datasets.SplitGenerator(
                    name=split._name,
                    gen_kwargs={
                        'data_file': dl_dirs[split],
                        'split': split,
                    }
                )
            )
        return splits

    def _generate_examples(self, data_file, split):
        # document_path = os.path.join(data_file, _DOCUMENT)
        # summary_path = os.path.join(data_file, _SUMMARY)

        features = self.config.available_features
        with ExitStack() as stack:
            files = [stack.enter_context(open(os.path.join(data_file, feature))) \
                     for feature in features]
            
            for idx, sample_data in enumerate(zip(*files)):
                yield idx, {
                    feature: feature_data  
                    for (feature, feature_data) in zip(features, sample_data)
                }
            
        # with open(document_path, 'r') as document_reader, open(summary_path, 'r') as summary_reader:
        #     for idx, (document, summary) in enumerate(zip(document_reader, summary_reader)):
        #         yield idx, {
        #             _DOCUMENT: document,
        #             _SUMMARY: summary,
        #         }