path: root/python/fatcat_tools/importers/arxiv.py

import sys
import json
import datetime
from bs4 import BeautifulSoup
from pylatexenc.latex2text import LatexNodes2Text

import fatcat_client
from .common import EntityImporter, clean
from .crossref import lookup_license_slug


latex2text = LatexNodes2Text()

def latex_to_text(raw):
    try:
        return latex2text.latex_to_text(raw).strip()
    except AttributeError:
        return raw.strip()

def parse_arxiv_authors(raw):
    if not raw:
        return []
    authors = raw.split(', ')
    if authors:
        last = authors[-1].split(" and ")
        if len(last) == 2:
            authors[-1] = last[0]
            authors.append(last[1])
    authors = [latex_to_text(a).strip() for a in authors]
    return authors

def test_parse_arxiv_authors():

    assert parse_arxiv_authors("Raphael Chetrite, Shamik Gupta, Izaak Neri and \\'Edgar Rold\\'an") == [
        "Raphael Chetrite",
        "Shamik Gupta",
        "Izaak Neri",
        "Édgar Roldán",
    ]
    assert parse_arxiv_authors("Izaak Neri and \\'Edgar Rold\\'an") == [
        "Izaak Neri",
        "Édgar Roldán",
    ]
    assert parse_arxiv_authors("Raphael Chetrite Shamik Gupta") == [
        "Raphael Chetrite Shamik Gupta",
    ]


class ArxivRawImporter(EntityImporter):
    """
    Converts arxiv.org "arXivRaw" OAI-PMH XML records to fatcat release entities

    TODO: arxiv_id lookup in API (rust) with no version specified should select
          the "most recent" version; can be a simple sort?
    """

    def __init__(self, api, **kwargs):

        eg_desc = kwargs.get('editgroup_description',
            "Automated import of arxiv metadata via arXivRaw OAI-PMH feed")
        eg_extra = kwargs.get('editgroup_extra', dict())
        eg_extra['agent'] = eg_extra.get('agent', 'fatcat_tools.ArxivRawImporter')
        # lower batch size, because multiple versions per entry (guessing 2-3 on average?)
        batch_size = kwargs.get('edit_batch_size', 50)
        super().__init__(api,
            editgroup_description=eg_desc,
            editgroup_extra=eg_extra,
            batch_size=batch_size,
            **kwargs)
        self._test_override = False


    def parse_record(self, record):

        if not record:
            return None
        metadata = record.arXivRaw
        if not metadata:
            return None
        extra = dict()
        extra_arxiv = dict()

        base_id = metadata.id.string
        doi = None
        if metadata.doi and metadata.doi.string:
            doi = metadata.doi.string.lower().strip()
            assert doi.startswith('10.')
        title = latex_to_text(metadata.title.string)
        authors = parse_arxiv_authors(metadata.authors.string)
        contribs = [fatcat_client.ReleaseContrib(raw_name=a, role='author') for a in authors]

        lang = "en"     # the vast majority in english
        if metadata.comments and metadata.comments.string:
            comments = metadata.comments.string.strip()
            extra_arxiv['comments'] = comments
            if 'in french' in comments.lower():
                lang = 'fr'
            elif 'in spanish' in comments.lower():
                lang = 'es'
            elif 'in portuguese' in comments.lower():
                lang = 'pt'
            elif 'in hindi' in comments.lower():
                lang = 'hi'
            elif 'in japanese' in comments.lower():
                lang = 'ja'
            elif 'in german' in comments.lower():
                lang = 'de'
            elif 'simplified chinese' in comments.lower():
                lang = 'zh'
            elif 'in russian' in comments.lower():
                lang = 'ru'
            # more languages?

        # don't know!
        release_type = "article"
        number = None

        if metadata.find('journal-ref') and metadata.find('journal-ref').string:
            journal_ref = metadata.find('journal-ref').string.strip()
            extra_arxiv['journal_ref'] = journal_ref
            if "conf." in journal_ref.lower() or "proc." in journal_ref.lower():
                release_type = "conference-paper"
        if metadata.find('report-no') and metadata.find('report-no').string:
            number = metadata.find('report-no').string.strip()
            release_type = "report"
        if metadata.find('acm-class') and metadata.find('acm-class').string:
            extra_arxiv['acm_class'] = metadata.find('acm-class').string.strip()
        if metadata.categories and metadata.categories.string:
            extra_arxiv['categories'] = metadata.categories.string.split()
        license_slug = None
        if metadata.license and metadata.license.string:
            license_slug = lookup_license_slug(metadata.license.string)
        abstracts = None
        if metadata.abstract:
            # TODO: test for this multi-abstract code path
            abstracts = []
            abst = metadata.abstract.string.strip()
            orig = None
            if '-----' in abst:
                both = abst.split('-----')
                abst = both[0].strip()
                orig = both[1].strip()
            if '$' in abst or '{' in abst:
                mime = "application/x-latex"
                abst_plain = latex_to_text(abst)
                abstracts.append(fatcat_client.ReleaseAbstract(content=abst_plain, mimetype="text/plain", lang="en"))
            else:
                mime = "text/plain"
            abstracts.append(fatcat_client.ReleaseAbstract(content=abst, mimetype=mime, lang="en"))
            if orig:
                abstracts.append(fatcat_client.ReleaseAbstract(content=orig, mimetype=mime))
                # indicates that fulltext probably isn't english either
                if lang == 'en':
                    lang = None


        # extra:
        #   withdrawn_date
        #   translation_of
        #   subtitle
        #   aliases
        #   container_name
        #   group-title
        #   arxiv: comments, categories, etc
        extra_arxiv['base_id'] = base_id
        extra['arxiv'] = extra_arxiv

        versions = []
        for version in metadata.find_all('version'):
            arxiv_id = base_id + version['version']
            release_date = version.date.string.strip()
            release_date = datetime.datetime.strptime(release_date, "%a, %d %b %Y %H:%M:%S %Z").date()
            # TODO: source_type?
            versions.append(fatcat_client.ReleaseEntity(
                work_id=None,
                title=title,
                #original_title
                version=version['version'],
                release_type="article",
                release_stage='submitted',
                release_date=release_date.isoformat(),
                release_year=release_date.year,
                ext_ids=fatcat_client.ReleaseExtIds(
                    arxiv=arxiv_id,
                ),
                number=number,
                language=lang,
                license_slug=license_slug,
                abstracts=abstracts,
                contribs=contribs,
                extra=extra,
            ))
        # TODO: assert that versions are actually in order
        assert versions

        # only apply DOI to most recent version (HACK)
        if doi:
            versions[-1].ext_ids.doi = doi
            versions[-1].release_stage = "published"
        return versions

    def try_update(self, versions):
        """
        This is pretty complex! There is no batch/bezerk mode for arxiv importer.

        For each version, do a lookup by full arxiv_id, and store work/release
        id results.
        
        If a version has a DOI, also do a doi lookup and store that result. If
        there is an existing release with both matching, set that as the
        existing work. If they don't match, use the full arxiv_id match and
        move on (maybe log or at least count the error?). If it's a
        one/or/other case, update the existing release (and mark version as
        existing).

        If there was any existing release, take its work_id.

        Iterate back through versions. If it didn't already exist, insert it
        with any existing work_id. If there wasn't an existing work_id, lookup
        the new release (by rev from edit?) and use that for the rest.

        Do not pass any versions on for batch insert.
        """

        # first do lookups
        any_work_id = None
        for v in versions:
            v._existing_work_id = None
            v._updated = False
            existing = None
            existing_doi = None
            try:
                existing = self.api.lookup_release(arxiv=v.ext_ids.arxiv)
            except fatcat_client.rest.ApiException as err:
                if err.status != 404:
                    raise err
            if v.ext_ids.doi:
                try:
                    existing_doi = self.api.lookup_release(arxiv=v.ext_ids.arxiv)
                except fatcat_client.rest.ApiException as err:
                    if err.status != 404:
                        raise err
            if existing_doi:
                if existing and existing.ident == existing_doi.ident:
                    # great, they match and have idents, nothing to do
                    pass
                elif existing and existing.ident != existing_doi.ident:
                    # could be bad, or could be that a new arxiv version was
                    # created (update?)
                    # stick with arxiv_id match as existing; don't update anything
                    pass
                else:
                    assert not existing
                    if not existing_doi.ext_ids.arxiv_id:
                        # update the existing DOI-based record with our full arxiv_id
                        existing_doi.ext_ids.arxiv_id = v.ext_ids.arxiv_id
                        self.api.update_release(self.get_editgroup_id(), existing_doi.ident, existing_doi)
                        self.counts['update'] += 1
                        # as a flag to not count below
                        v._updated = True
                    existing = existing_doi

            if existing:
                v._existing_work_id = existing.work_id
                any_work_id = existing.work_id

        last_edit = None
        for v in versions:
            if v._existing_work_id:
                if not v._updated:
                    self.counts['exists'] += 1
                continue
            if not any_work_id and last_edit:
                # fetch the last inserted release from this group
                r = self.api.get_release_revision(last_edit.revision)
                assert r.work_id
                any_work_id = r.work_id
            v.work_id = any_work_id
            last_edit = self.api.create_release(self.get_editgroup_id(), v)
            self.counts['insert'] += 1

        return False

    def insert_batch(self, batch_batch):
        # there is no batch/bezerk mode for arxiv importer, except for testing
        if self._test_override:
            for batch in batch_batch:
                self.api.create_release_auto_batch(fatcat_client.ReleaseAutoBatch(
                    editgroup=fatcat_client.Editgroup(
                        description=self.editgroup_description,
                        extra=self.editgroup_extra),
                    entity_list=batch))
                self.counts['insert'] += len(batch) - 1
        else:
            raise NotImplementedError()

    def parse_file(self, handle):

        # 1. open with beautiful soup
        soup = BeautifulSoup(handle, "xml")

        # 2. iterate over articles, call parse_article on each
        for article in soup.find_all("record"):
            resp = self.parse_record(article)
            print(json.dumps(resp))
            #sys.exit(-1)

if __name__=='__main__':
    parser = ArxivRawImporter(None)
    parser.parse_file(open(sys.argv[1]))