From 16256f8ed119c072c09b13b0b1a6d4a56bed5113 Mon Sep 17 00:00:00 2001 From: Bryan Newbold Date: Fri, 25 Jan 2019 16:43:18 -0800 Subject: improved journal metadata munger --- extra/journal_metadata/README.md | 18 +- extra/journal_metadata/parse_merge_metadata.py | 407 +++++++++++++++++++------ 2 files changed, 325 insertions(+), 100 deletions(-) (limited to 'extra/journal_metadata') diff --git a/extra/journal_metadata/README.md b/extra/journal_metadata/README.md index 61dbc6b0..dec32624 100644 --- a/extra/journal_metadata/README.md +++ b/extra/journal_metadata/README.md @@ -17,7 +17,7 @@ A few sources of normalization/mappings: - Original: - Snapshot: - ISO 639-1 language codes: https://datahub.io/core/language-codes -- ISO 3166-1 alpha-2 nation codes +- ISO 3166-1 alpha-2 country codes In order of precedence (first higher than later): @@ -60,6 +60,22 @@ ISSN-L, then write out to disk as JSON. Then the journal-metadata importer takes a subset of fields and inserts to fatcat. Lastly, the elasticsearch transformer takes a subset/combination of +## ISSN-L Munging + +Unfortunately, there seem to be plenty of legitimate ISSNs that don't end up in +the ISSN-L table. On the portal.issn.org public site, these are listed as: + + "This provisional record has been produced before publication of the + resource. The published resource has not yet been checked by the ISSN + Network.It is only available to subscribing users." + +For example: + +- 2199-3246/2199-3254: Digital Experiences in Mathematics Education + +Instead of just dropping these entirely, we're currently munging these by +putting the electronic or print ISSN in the ISSN-L position. + ## Python Helpers/Libraries - ftfy diff --git a/extra/journal_metadata/parse_merge_metadata.py b/extra/journal_metadata/parse_merge_metadata.py index c9909a8e..00802458 100755 --- a/extra/journal_metadata/parse_merge_metadata.py +++ b/extra/journal_metadata/parse_merge_metadata.py @@ -3,6 +3,7 @@ import sys, csv, json import ftfy import pycountry +from collections import Counter ISSNL_FILE = 'data/20181203.ISSN-to-ISSN-L.txt' @@ -23,6 +24,102 @@ JSTOR_FILE = 'data/jstor_all-archive-titles.txt' SIM_FILE = 'data/MASTER TITLE_METADATA_LIST_20171019.converted.csv' IA_CRAWL_FILE = 'data/journal_homepage_results.partial.tsv' +# NOTE: this is a partial list, focusing on non-publisher hosted platforms and +# software frameworks +PLATFORM_MAP = { + 'OJS': 'ojs', + 'BMC': 'bmc', + 'SciELO Brazil': 'scielo', + 'SciELO Argentina': 'scielo', + 'SciELO': 'scielo', + 'SciELO Mexico': 'scielo', + 'SciELO Spain': 'scielo', + 'SciELO Portugal': 'scielo', + 'WordPress': 'wordpress', + 'Sciendo': 'sciendo', + 'Drupal': 'drupal', + 'revues.org': 'openedition', +} + +MIMETYPE_MAP = { + 'PDF': 'application/pdf', + 'HTML': 'text/html', + 'XML': 'application/xml', +} + +def unquote(s): + if s.startswith('"'): + s = s[1:] + if s.endswith('"'): + s = s[:-1] + if s.endswith('.'): + s = s[:-1] + return s.strip() + +def parse_lang(s): + if not s or s in ('Not applicable', 'Multiple languages', 'Unknown'): + return None + try: + if len(s) == 2: + lang = pycountry.languages.get(alpha2=s.lower()) + elif len(s) == 3: + lang = pycountry.languages.get(alpha3=s.lower()) + else: + lang = pycountry.languages.get(name=s) + return lang.alpha2.lower() + except KeyError: + return None + except AttributeError: + return None + +def parse_country(s): + if not s or s in ('Unknown'): + return None + try: + if len(s) == 2: + country = pycountry.countries.get(alpha2=s.lower()) + else: + country = pycountry.countries.get(name=s) + except KeyError: + return None + return country.alpha2.lower() + +def gaps_to_spans(first, last, gaps): + if not gaps: + return [[first, last]] + if not (last >= first and max(gaps) < last and min(gaps) > first): + # mangled + print("mangled years: {}".format((first, last, gaps))) + return [] + full = list(range(first, last+1)) + for missing in gaps: + full.remove(missing) + spans = [] + low = None + last = None + for year in full: + if not low: + low = year + last = year + continue + if year != last+1: + spans.append([low, last]) + low = year + last = year + last = year + if low: + spans.append([low, last]) + return spans + +def test_gaps(): + assert gaps_to_spans(1900, 1900, None) == \ + [[1900, 1900]] + assert gaps_to_spans(1900, 1903, None) == \ + [[1900, 1903]] + assert gaps_to_spans(1900, 1902, [1901]) == \ + [[1900, 1900], [1902, 1902]] + assert gaps_to_spans(1950, 1970, [1955, 1956, 1965]) == \ + [[1950, 1954], [1957, 1964], [1966, 1970]] class Munger(): """ @@ -33,8 +130,8 @@ class Munger(): first_year: year (integer) last_year: if publishing has stopped languages: array of ISO codes; first is the "primary" language - nation: ISO shortcode of nation published from - url: homepage + country: ISO shortcode of country published from + urls: homepage links abbrev: string default_license: slug original_name: native name (if name is translated) @@ -52,12 +149,8 @@ class Munger(): field: TODO: narrower description of field subjects: TODO? - TODO: more ftfy? - TODO: remove surrounding quotes - TODO: null ISSN-L? - TODO: sherpa OA: 'Paid OA options' or 'All journals OA' - TODO: mailto: in urls - TODO: empty gaps (sim) + TODO: so many missing ISSN/ISSN-L + TODO: abbrev """ def __init__(self): @@ -66,11 +159,11 @@ class Munger(): self.read_issn_map_file(f) def run(self, out_path): - self.load_road(ROAD_FILE) self.load_doaj(DOAJ_FILE) - self.load_crossref(CROSSREF_FILE) self.load_norwegian(NORWEGIAN_FILE) + self.load_crossref(CROSSREF_FILE) self.load_sherpa_romeo(SHERPA_ROMEO_JOURNAL_FILE, SHERPA_ROMEO_POLICY_FILE) + self.load_road(ROAD_FILE) self.load_kbart('lockss', LOCKSS_FILE) self.load_kbart('clockss', CLOCKSS_FILE) self.load_kbart('portico', PORTICO_FILE) @@ -110,99 +203,195 @@ class Munger(): def add_issn(self, raw_issn=None, issne=None, issnp=None, name=None, publisher=None): # do ISSN => ISSN-L mappings for any raw ISSNs - lookup = raw_issn or issne or issnp - lookup = lookup.strip() - if not (len(lookup) == 9 and lookup[4] == '-'): - print(lookup) - print(len(lookup)) - print(lookup[4]) - assert len(lookup) == 9 and lookup[4] == '-' - issnl = self.issn2issnl(lookup.upper()) + issnl = None + if not (raw_issn or issne or issnp): + return None, 'no-issn' + for lookup in (issnp, issne, raw_issn): + if not lookup: + continue + lookup = lookup.strip().upper() + #if not (len(lookup) == 9 and lookup[4] == '-'): + # print(lookup) + # print(len(lookup)) + # print(lookup[4]) + # return None, 'invalid-issn' + #assert len(lookup) == 9 and lookup[4] == '-' + issnl = self.issn2issnl(lookup) + if issnl: + break + if not issnl: + #print((raw_issn, issne, issnp)) + # UGH. + issnl = issne or issnp or raw_issn + if not issnl: + return None, 'no-issnl' + issnl = issnl.strip().upper() + assert len(issnl) == 9 and issnl[4] == '-' + status = 'found-munge' + else: + status = 'found' # lookup ISSN-Ls in data (or create one) if not issnl in self.data: + status = 'created' self.data[issnl] = dict(issnl=issnl) d = self.data[issnl] # if name/publisher not set, do so if name and not 'name' in d: - self.data[issnl]['name'] = ftfy.fix_text(name).strip() + name = unquote(ftfy.fix_text(name)) + if name: + self.data[issnl]['name'] = name if publisher and not 'publisher' in d: - self.data[issnl]['publisher'] = ftfy.fix_text(publisher).strip() + publisher = unquote(ftfy.fix_text(publisher)) + if publisher: + self.data[issnl]['publisher'] = publisher if issne and not 'issne' in d: self.data[issnl]['issne'] = issne if issnp and not 'issnp' in d: self.data[issnl]['issnp'] = issnp # always return ISSN-L - return issnl + return issnl, status + + def add_lang(self, issnl, lang): + if not (lang and issnl): + return + lang = parse_lang(lang) + if not lang: + return + if 'languages' not in self.data[issnl]: + self.data[issnl]['languages'] = [lang] + elif lang not in self.data[issnl]['languages']: + self.data[issnl]['languages'].append(lang) + + def add_url(self, issnl, url): + if not (url and issnl) or 'mailto:' in url.lower() or url in ('http://n/a', 'http://N/A'): + return + if url.startswith('www.'): + url = "http://" + url + url.replace('Http://', 'http://') + if 'urls' not in self.data[issnl]: + self.data[issnl]['urls'] = [url] + elif url not in self.data[issnl]['urls']: + self.data[issnl]['urls'].append(url) + + def add_country(self, issnl, country): + if not (country and issnl): + return + country = parse_country(country) + if not country: + return + if 'country' not in self.data[issnl]: + self.data[issnl]['country'] = country + + def add_mimetype(self, issnl, val): + if not (val and issnl): + return + mimetype = None + if '/' in val: + mimetype = val + else: + mimetype = MIMETYPE_MAP.get(val) + if not mimetype: + return + if 'mimetypes' not in self.data[issnl]: + self.data[issnl]['mimestypes'] = [mimetype] + elif mimetype not in self.data[issnl]['mimestypes']: + self.data[issnl]['mimestypes'].append(mimetype) def load_entrez(self, path): print("##### Loading Entrez...") # JrId,JournalTitle,MedAbbr,"ISSN (Print)","ISSN (Online)",IsoAbbr,NlmId reader = csv.DictReader(open(path)) - skipped = 0 - count = 0 + counts = Counter() for row in reader: if not (row.get('ISSN (Online)') or row.get('ISSN (Print)')): - skipped += 1 + counts['skipped'] += 1 continue - issnl = self.add_issn( + issnl, status = self.add_issn( issne=row.get('ISSN (Online)'), issnp=row.get('ISSN (Print)'), name=row['JournalTitle'], ) - count += 1 - print("Matched {}".format(count)) - print("Skipped {} for not having ISSNs".format(skipped)) + if row['IsoAbbr'] and not 'abbrev' in self.data[issnl]: + self.data[issnl]['abbrev'] = row['IsoAbbr'].strip() + counts[status] += 1 + print(counts) def load_road(self, path): print("##### Loading ROAD...") reader = csv.DictReader(open(path), delimiter='\t', fieldnames=("ISSN", "ISSN-L", "Short Title", "Title", "Publisher", "URL1", "URL2", "Region", "Lang1", "Lang2") ) - count = 0 + counts = Counter() for row in reader: - issnl = self.add_issn( + issnl, status = self.add_issn( raw_issn=row['ISSN-L'], name=row['Short Title'], publisher=row['Publisher'], ) - count += 1 + counts[status] += 1 + if not issnl: + continue d = self.data[issnl] - if row['URL1'] and not 'url' in d: - self.data[issnl]['url'] = row['URL1'] + if row['URL1']: + self.add_url(issnl, row['URL1']) + if row['URL2']: + self.add_url(issnl, row['URL2']) + if row['Lang1']: + self.add_lang(issnl, row['Lang1']) + if row['Lang2']: + self.add_lang(issnl, row['Lang2']) # TODO: region mapping: "Europe and North America" # TODO: lang mapping: already alpha-3 self.data[issnl]['road'] = dict(as_of=ROAD_DATE) - print("Matched {}".format(count)) + print(counts) def load_doaj(self, path): print("##### Loading DOAJ...") - #Journal title Journal URL Alternative title ISSN-print ISSN-electronic Publisher Society or institution Platform, host or aggregator Country of publisher Journal article processing charges (APCs) ... Deposit policy directory Author holds copyright without restrictions Copyright information URL Author holds publishing rights without restrictions Publishing rights information URL DOAJ Seal Tick: Accepted after March 2014 Added on Date Subjects ISSN-L + #Journal title,Journal URL,Alternative title,Journal ISSN (print version),Journal EISSN (online version),Publisher,Society or institution,"Platform, host or aggregator",Country of publisher,Journal article processing charges (APCs),APC information URL,APC amount,Currency,Journal article submission fee,Submission fee URL,Submission fee amount,Submission fee currency,Number of articles publish in the last calendar year,Number of articles information URL,Journal waiver policy (for developing country authors etc),Waiver policy information URL,Digital archiving policy or program(s),Archiving: national library,Archiving: other,Archiving infomation URL,Journal full-text crawl permission,Permanent article identifiers,Journal provides download statistics,Download statistics information URL,First calendar year journal provided online Open Access content,Full text formats,Keywords,Full text language,URL for the Editorial Board page,Review process,Review process information URL,URL for journal's aims & scope,URL for journal's instructions for authors,Journal plagiarism screening policy,Plagiarism information URL,Average number of weeks between submission and publication,URL for journal's Open Access statement,Machine-readable CC licensing information embedded or displayed in articles,URL to an example page with embedded licensing information,Journal license,License attributes,URL for license terms,Does this journal allow unrestricted reuse in compliance with BOAI?,Deposit policy directory,Author holds copyright without restrictions,Copyright information URL,Author holds publishing rights without restrictions,Publishing rights information URL,DOAJ Seal,Tick: Accepted after March 2014,Added on Date,Subjects reader = csv.DictReader(open(path)) - count = 0 + counts = Counter() for row in reader: - issnl = self.add_issn( + issnl, status = self.add_issn( issnp=row['Journal ISSN (print version)'], issne=row['Journal EISSN (online version)'], name=row['Journal title'], publisher=row['Publisher'], ) - count += 1 + counts[status] += 1 + if not issnl: + continue d = self.data[issnl] doaj = dict(as_of=DOAJ_DATE) # TODO: work_level: bool (are work-level publications deposited with DOAJ?) - # TODO: archiving: array, can include 'library' or 'other' - if row['Platform, host or aggregator']: - # TODO: mapping here? - self.data[issnl]['platform'] = row['Platform, host or aggregator'] + if row['Digital archiving policy or program(s)']: + doaj['archive'] = [a.strip() for a in row['Digital archiving policy or program(s)'].split(',') if a.strip()] + elif row['Archiving: national library']: + doaj['archive'] = ['national-library'] + + crawl_permission = row['Journal full-text crawl permission'] + if crawl_permission: + doaj['crawl-permission'] = dict(Yes=True, No=False)[crawl_permission] + # TODO: Permanent article identifiers + default_license = row['Journal license'] + if default_license and default_license.startswith('CC'): + self.data[issnl]['default_license'] = default_license.replace('CC ', 'CC-').strip() + + self.add_mimetype(issnl, row['Full text formats']) + platform = PLATFORM_MAP.get(row['Platform, host or aggregator']) + if platform: + self.data[issnl]['platform'] = platform if row['DOAJ Seal']: doaj['seal'] = {"no": False, "yes": True}[row['DOAJ Seal'].lower()] if row['Country of publisher']: - # TODO: country mapping - self.data[issnl]['country'] = row['Country of publisher'] + self.add_country(issnl, row['Country of publisher']) + if row['Full text language']: + self.add_lang(issnl, row['Full text language']) + if row['Journal URL']: + self.add_url(issnl, row['Journal URL']) # TODO: Subjects self.data[issnl]['doaj'] = doaj - print("Matched {}".format(count)) + print(counts) def load_sherpa_romeo(self, journal_path, policy_path): # first load policies @@ -219,36 +408,35 @@ class Munger(): raw_file = open(journal_path, 'rb').read().decode(errors='replace') fixed_file = ftfy.fix_text(raw_file) reader = csv.DictReader(fixed_file.split('\n')) - count = 0 + counts = Counter() for row in reader: #row['Journal Title'] = row.pop('\ufeffJournal Title') row.update(policies[row['RoMEO Record ID']]) - issnl = self.add_issn( + issnl, status = self.add_issn( issnp=row['ISSN'], issne=row['ESSN'], name=row['Journal Title'], publisher=row['Publisher'], ) - count += 1 + counts[status] += 1 + if not issnl: + continue d = self.data[issnl] sherpa_romeo = dict() if row['RoMEO colour']: sherpa_romeo['color'] = row['RoMEO colour'] - if row['Open Access Publishing']: - # TODO: boolean? - sherpa_romeo['oa'] = row['Open Access Publishing'] - if row['Country'] and not 'country' in d: - self.data[issnl]['country'] = row['Country'].lower() + # row['Open Access Publishing'] + if row['Country']: + self.add_country(issnl, row['Country']) self.data[issnl]['sherpa_romeo'] = sherpa_romeo - print("Matched {}".format(count)) + print(counts) def load_norwegian(self, path): print("##### Loading Norwegian Registry...") #pandas.read_csv(NORWEGIAN_FILE, sep=';', encoding="ISO-8859-1") #NSD tidsskrift_id;Original title;International title;Present Level (2018);Print ISSN;Online ISSN;Open Access;NPI Scientific Field;NPI Academic Discipline;URL;Publishing Company;Publisher;Country of publication;Language;Level 2019;Level 2018;Level 2017;Level 2016;Level 2015;Level 2014;Level 2013;Level 2012;Level 2011;Level 2010;Level 2009;Level 2008;Level 2007;Level 2006;Level 2005;Level 2004;itar_id reader = csv.DictReader(open(path, encoding="ISO-8859-1"), delimiter=";") - count = 0 - skip = 0 + counts = Counter() for row in reader: issnp = row['Print ISSN'] issne = row['Online ISSN'] @@ -257,15 +445,17 @@ class Munger(): if issnp and len(issnp.strip()) != 9: issnp = None if not (issnp or issne): - skip += 1 + counts['no-issn'] += 1 continue - issnl = self.add_issn( + issnl, status = self.add_issn( issnp=issnp, issne=issne, name=row['International title'], publisher=row['Publisher'], ) - count += 1 + counts[status] += 1 + if not issnl: + continue d = self.data[issnl] norwegian = dict(as_of=NORWEGIAN_DATE) norwegian['level'] = int(row['Present Level (2018)']) @@ -273,15 +463,14 @@ class Munger(): if row['Original title'] != row['International title'] and not 'original_name' in d: self.data[issnl]['original_name'] = row['Original title'] - if row['Country of publication'] and not 'country' in d: - # TODO: country mapping - self.data[issnl]['country'] = row['Country of publication'] - if row['Language'] and not 'language' in d: - # TODO: language mapping - self.data[issnl]['language'] = row['Language'] + if row['Country of publication']: + self.add_country(issnl, row['Country of publication']) + if row['Language']: + self.add_lang(issnl, row['Language']) + if row['URL']: + self.add_url(issnl, row['URL']) self.data[issnl]['norwegian'] = norwegian - print("Skipped {} for mangled ISSN".format(skip)) - print("Matched {}".format(count)) + print(counts) def load_kbart(self, name, path): print("##### Loading KBART file for {}...".format(name)) @@ -289,98 +478,119 @@ class Munger(): raw_file = open(path, 'rb').read().decode(errors='replace') fixed_file = ftfy.fix_text(raw_file) reader = csv.DictReader(fixed_file.split('\n'), delimiter='\t') - count = 0 - skip = 0 + counts = Counter() for row in reader: if not row['print_identifier'] and not row['online_identifier']: - skip += 1 + counts['no-issn'] += 1 continue - issnl = self.add_issn( + issnl, status = self.add_issn( issnp=row['print_identifier'], issne=row['online_identifier'], name=row['publication_title'], publisher=row['publisher_name'], ) - count += 1 + counts[status] += 1 + if not issnl: + continue d = self.data[issnl] if not 'kbart' in d: self.data[issnl]['kbart'] = dict() kbart = dict() if row['date_first_issue_online'] and row['date_last_issue_online']: - kbart['year_span'] = [[int(row['date_first_issue_online'][:4]), int(row['date_last_issue_online'][:4])]] + kbart['year_spans'] = [[int(row['date_first_issue_online'][:4]), int(row['date_last_issue_online'][:4])]] self.data[issnl]['kbart'][name] = kbart - print("Skipped {} missing ISSN".format(skip)) - print("Matched {}".format(count)) + print(counts) def load_crossref(self, path): print("##### Loading Crossref...") #"JournalTitle","JournalID","Publisher","pissn","eissn","additionalIssns","doi","(year1)[volume1]issue1,issue2,issue3(year2)[volume2]issue4,issues5" reader = csv.DictReader(open(path)) - count = 0 - skip = 0 + counts = Counter() for row in reader: if row['pissn'] and len(row['pissn']) == 8: row['pissn'] = row['pissn'][:4] + '-' + row['pissn'][4:] if row['eissn'] and len(row['eissn']) == 8: row['eissn'] = row['eissn'][:4] + '-' + row['eissn'][4:] if not (row['pissn'] or row['eissn']): - skip += 1 + counts['no-issn'] += 1 continue - issnl = self.add_issn( + issnl, status = self.add_issn( issnp=row['pissn'], issne=row['eissn'], name=row['JournalTitle'], publisher=row['Publisher'], ) - count += 1 + counts[status] += 1 + if not issnl: + continue d = self.data[issnl] crossref = dict() if row['doi']: crossref['doi'] = row['doi'] + crossref['any'] = True self.data[issnl]['crossref'] = crossref - print("Skipped {} missing ISSN".format(skip)) - print("Matched {}".format(count)) + print(counts) def load_sim(self, path): print("##### Loading SIM Metadata...") #NA Pub Cat ID,Title,Publisher,ISSN,Impact Rank,Total Cities,Journal Impact Factor,Eigenfact or Score,First Volume,Last Volume,NA Gaps,"Scholarly / Peer-\n Reviewed","Peer-\n Reviewed",Pub Type,Pub Language,Subjects reader = csv.DictReader(open(path)) - count = 0 - skip = 0 + counts = Counter() for row in reader: if not row['ISSN'] or row['ISSN'] == "NULL": - skip += 1 + counts['no-issn'] += 1 continue - issnl = self.add_issn( + issnl, status = self.add_issn( raw_issn=row['ISSN'][:9], name=row['Title'], publisher=row['Publisher'], ) - count += 1 + counts[status] += 1 + if not issnl: + continue d = self.data[issnl] sim = dict() sim['id'] = row['NA Pub Cat ID'] - sim['first_year'] = row['First Volume'] - sim['last_year'] = row['Last Volume'] - sim['gaps'] = row['NA Gaps'] - # TODO: 'Pub Language' + first_year = row['First Volume'] + if first_year: + first_year = int(first_year) + sim['first_year'] = int(row['First Volume']) + else: + first_year = None + last_year = row['Last Volume'] + if last_year: + last_year = int(last_year) + sim['last_year'] = last_year + else: + last_year = None + gaps = [int(g) for g in row['NA Gaps'].split(';') if g.strip()] + if gaps: + sim['gaps'] = gaps + if first_year and last_year: + sim['year_spans'] = gaps_to_spans(first_year, last_year, gaps) + if row['Pub Language']: + self.add_lang(issnl, row['Pub Language']) # TODO: 'Pub Type' + all_keys = list(sim.keys()) + for k in all_keys: + if not sim[k]: + sim.pop(k) self.data[issnl]['sim'] = sim - print("Skipped {} missing ISSN".format(skip)) - print("Matched {}".format(count)) + print(counts) def load_homepage_crawl(self, path): print("##### Loading IA Homepage Crawl Results...") reader = csv.DictReader(open(path), delimiter='\t', fieldnames=("ISSN", "first_url", "first_status", "last_status", "last_url") ) - count = 0 - skip = 0 + counts = Counter() for row in reader: - issnl = self.add_issn( + issnl, status = self.add_issn( raw_issn=row['ISSN'], ) - count += 1 + counts[status] += 1 + if not issnl: + continue d = self.data[issnl] ia = d.get('ia', dict()) ia['homepage_status'] = int(row['last_status']) @@ -389,8 +599,7 @@ class Munger(): else: ia['homepage_url'] = row['first_url'] self.data[issnl]['ia'] = ia - print("Skipped {} missing ISSN".format(skip)) - print("Matched {}".format(count)) + print(counts) if __name__=='__main__': munger = Munger() -- cgit v1.2.3