EThOS-PhD-metadata.py

# coding=utf-8
"""British Library EThos dataset"""

import csv
from datetime import datetime
import datasets
from datasets.features import Features
from datasets.tasks import LanguageModeling

_CITATION = """\ 
@misc{british library_genre,
title={UK Doctoral Thesis Metadata from EThOS},
url={UK Doctoral Thesis Metadata from EThOS},
author={{British Library} and  {Rosie, Heather}},
year={2021}}
"""


_DESCRIPTION = """\
The data in this collection comprises the bibliographic metadata for all UK doctoral theses listed in EThOS, the UK's national thesis service.
We estimate the data covers around 98% of all PhDs ever awarded by UK Higher Education institutions, dating back to 1787.
Thesis metadata from every PhD-awarding university in the UK is included.
"""


_HOMEPAGE = "https://doi.org/10.23636/rcm4-zk44"
_LICENSE = "CC BY 4.0 Attribution"


_URL = "https://bl.iro.bl.uk/downloads/05b31c0e-da22-4b9f-a17c-35880aa111f4?locale=en"


features = Features(
    {
        "Title": datasets.Value("string"),
        "DOI": datasets.Value("string"),
        "Author": datasets.Value("string"),
        "Author ISNI": datasets.Value("string"),
        "ORCID": datasets.Value("string"),
        "Institution": datasets.Value("string"),
        "Institution ISNI": datasets.Value("string"),
        "Date": datasets.Value("timestamp[s]"),
        "Qualification": datasets.Value("string"),
        "Abstract": datasets.Value("string"),
        "Subject Discipline": datasets.ClassLabel(
            names=[
                "Physical Sciences",
                "Biological Sciences",
                "Engineering & Technology",
                "Mathematics & Statistics",
                "Agriculture & Veterinary Sciences",
                "Medicine & Health",
                "Computer Science",
                "Philosophy, Psychology & Religious Studies",
                "Business & Administrative Studies",
                "Education",
                "Language & Literature",
                "Social, Economic & Political Studies",
                "Architecture, Building & Planning",
                "History & Archaeology",
                "Creative Arts & Design",
                "Law",
                "Sport & Recreation",
                "Librarianship & Information Science",
                "Music",
                " ",
            ]
        ),
        "Supervisor(s)": datasets.Value("string"),
        "Funder(s)": datasets.Value("string"),
        "EThOS URL": datasets.Value("string"),
        "IR URL": datasets.Value("string"),
    }
)


class Ethos(datasets.GeneratorBasedBuilder):

    VERSION = datasets.Version("1.1.0")
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="all",
            version=VERSION,
            description="",
        ),
        datasets.BuilderConfig(
            "skip_empty_abstracts",
            version=VERSION,
            description="EThOs skipping entires with no abstract",
        ),
    ]

    DEFAULT_CONFIG_NAME = "all"

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            supervised_keys=None,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
            task_templates=[LanguageModeling(text_column="Abstract")],
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        data_file = dl_manager.download(_URL)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "filepath": data_file,
                },
            ),
        ]

    def _generate_examples(self, filepath):
        """Yields examples as (key, example) tuples."""
        if self.config.name == "skip_empty_abstracts":
            skip = True
        else:
            skip = False
        with open(filepath, encoding="latin-1") as f:
            reader = csv.DictReader(f)
            id_ = 0
            for row in reader:
                abstract = row["Abstract"]
                if skip and len(abstract) < 2:
                    continue
                try:
                    date = datetime.strptime(row["Date"], "%Y")
                except ValueError:
                    date = None
                id_ += 1
                yield id_, {
                    "Title": row["Title"],
                    "DOI": row["DOI"],
                    "Author": row["Author"],
                    "Author ISNI": row["Author ISNI"],
                    "ORCID": row["ORCID"],
                    "Institution": row["Institution"],
                    "Institution ISNI": row["Institution ISNI"],
                    "Date": date,
                    "Qualification": row["Qualification"],
                    "Abstract": abstract,
                    "Subject Discipline": row["Subject Discipline"],
                    "Supervisor(s)": row["Supervisor(s)"],
                    "Funder(s)": row["Funder(s)"],
                    "EThOS URL": row["EThOS URL"],
                    "IR URL": row["IR URL"],
                }