# Cargo datos
data = pickle.load(open(this_file, 'rb'))
progress_bar = ChargingBar("Evaluating file {} - {}/{}".format(
this_filename, this_file_idx + 1, len(test_filepaths)),
suffix="%(percent)d%%")
if realistic_check:
# Separo datos en pedazos
sign = [
data['data'][i:i + slice_samples]
for i in range(0, len(data['data']), slice_samples)
]
progress_bar.max = len(sign)
progress_bar.start()
for this_slice_idx, this_slice in enumerate(sign):
"""
if this_slice_idx % (round(len(sign) / 20)) == 0:
print("{}/{} slices evaluated.".format(this_slice_idx, len(sign)))
"""
progress_bar.next()
# Si no pongo este if, el ultimo pedazo podria ser de distinto tamaño y fallar
if len(this_slice) == slice_samples:
# Analisis
this_data = torch.Tensor(this_slice)
this_data = this_data.view(1, 1,
def scrape(self, term):
"""
Scrapes metadata of PubMed articles returned by search term query, processes
abstracts, and stores relevant articles
:param term: PubMed term query
"""
print(
f'Collection: {self._collection.database.name}.{self._collection.name}. Database: PubMed. Term: {term}.'
)
base = 'https://eutils.ncbi.nlm.nih.gov/entrez/eutils'
retmax = 10000
no_id = 0
unreadable = 0
abstracts = []
articles = []
page = 0
total = retmax
# progress bar
bar = ChargingBar('Getting metadata:',
max=total,
suffix='%(index)d of %(max)d - %(elapsed_td)s')
while page < total:
# gets and stores to history UIDs of query
url = f'{base}/esearch.fcgi?db=pubmed&term={term}&retstart={page}'
url += f'&retmax={retmax}&usehistory=y&api_key={PUBMED_API_KEY}'
response = requests.get(url)
if not response.ok:
print(
f'\nPubmedScraper could not get UIDs for \'{term}\' on page {page}.'
)
continue
# gets info for retrieving UIDs from history
soup = BeautifulSoup(response.content, 'html.parser')
web = soup.webenv.string
key = soup.querykey.string
total = int(soup.count.string)
bar.max = total
# gets metadata for articles from UIDs
url = f'{base}/efetch.fcgi?db=pubmed&WebEnv={web}'
url += f'&query_key={key}&retstart={page}&retmax={retmax}'
url += f'&retmode=xml&api_key={PUBMED_API_KEY}'
response = requests.get(url)
if not response.ok:
print(
f'\nPubmedScraper could not get metadata for \'{term}\' on page {page}.'
)
continue
soup = BeautifulSoup(response.content, 'html.parser')
entries = soup.find_all('pubmedarticle')
for article in entries:
# ignore abstract if doi and uid are null
doi = self._get_string(
article.find('elocationid', eidtype='doi'))
uid = self._get_string(article.find('pmid'))
if not doi and not uid:
no_id += 1
bar.next()
continue
# store abstract text for use by mat2vec below
abstract = self._remove_html(article.abstracttext)
# continues if paper does not have abstract
if not abstract:
unreadable += 1
bar.next()
continue
# segments abstract by sentence
doc = self.nlp(abstract)
sentences = []
is_unreadable = False
for sent in doc.sents:
# processes sentence text using processor from mat2vec
try:
tokens, materials = self.processor.process(sent.text)
except OverflowError:
is_unreadable = True
break
processed_sent = ' '.join(tokens)
sentences.append(processed_sent)
# if processor (from above) throws an error, skip the paper
if is_unreadable:
bar.next()
unreadable += 1
continue
processed_abstract = '\n'.join(sentences)
article = {
'doi':
doi,
'uid':
uid,
'title':
self._remove_html(article.articletitle),
'abstract':
abstract,
'creators':
self._get_authors(article.find_all('author')),
'publication_name':
self._remove_html(article.journal.title),
'issn':
self._get_string(article.find('issn', issntype='Print')),
'eissn':
self._get_string(
article.find('issn', issntype='Electronic')),
'publication_date':
self._get_date(article.articledate),
'database':
'pubmed',
'processed_abstract':
processed_abstract,
}
articles.append(article)
abstracts.append(processed_abstract)
bar.next()
# classify abstracts if 20000 have been stored
if len(abstracts) == 20000:
self._store(articles, abstracts)
articles = []
abstracts = []
page += retmax
bar.finish()
# unreadable papers
print(f'No DOI/UID: {no_id}')
print(f'Unreadable papers: {unreadable}')
# classifies and stores metadata
if abstracts:
self._store(articles, abstracts)
print()
else:
print('No abstracts to classify.\n')
return
# prints classifier metrics
for classifier in self._classifiers:
classifier.print_metrics()
classifier.reset_metrics()
# prints general tag metrics
if self._save:
print(f'Total articles analyzed: {self._gen_total}.')
print(
f'Stored {self._gen_new} new abstracts to \'{self._gen_tag}\'.'
)
print()
self._gen_new = 0
self._gen_total = 0
def scrape_faster(self, query):
"""
Note: requires institutional access by VPN, or else an error will be thrown
Faster implementation of scrape()
Scrapes metadata of Elsevier (ScienceDirect) articles returned
by query, processes abstracts, and stores relevant articles
:param query: Elsevier database query
"""
print(
f'Collection: {self._collection.database.name}.{self._collection.name}. Database: Elsevier. Query: {query}.'
)
# create url
url = f'https://api.elsevier.com/content/metadata/article?query=KEY({query})&apiKey={ELSEVIER_API_KEY}&httpAccept=application%2Fjson'
articles = []
abstracts = []
no_doi = 0
unreadable = 0
item = 0
total = 5000
# progress bar
bar = ChargingBar('Getting metadata:',
max=total,
suffix='%(index)d of %(max)d - %(elapsed_td)s')
while item < total:
response = requests.get(url)
if response.ok:
data = json.loads(response.content)['search-results']
records = data['entry']
# updates total to total number of papers in query
if item == 0:
total = min(5000, int(data['opensearch:totalResults']))
bar.max = total
# if there are no results, exit
if total == 0:
print('Search returned no results.\n')
return
for record in records:
doi = record.get('prism:doi')
if not doi:
no_doi += 1
bar.next()
continue
abstract = record.get('prism:teaser')
# if there is no abstract, skip this article
if not abstract:
unreadable += 1
bar.next()
continue
# segments abstract by sentence
doc = self.nlp(abstract)
sentences = []
is_unreadable = False
# processes sentence text using mat2vec processor
for sent in doc.sents:
try:
tokens, materials = self.processor.process(
sent.text)
except OverflowError:
is_unreadable = True
break
processed_sent = ' '.join(tokens)
sentences.append(processed_sent)
# if processor (from above) throws an error, skip the paper
if is_unreadable:
unreadable += 1
bar.next()
continue
processed_abstract = '\n'.join(sentences)
# create new document and store new article document if not in collection
article = {
'doi':
doi,
'uid':
None,
'title':
record.get('dc:title'),
'abstract':
abstract,
'url':
record.get('prism:url'),
'creators':
self._get_creators(data.get('dc:creator')),
'publication_name':
data.get('prism:publicationName'),
'issn':
record.get('prism:issn'),
'publication_date':
self._get_date(data.get('prism:coverDate')),
'database':
'elsevier',
'processed_abstract':
processed_abstract
}
articles.append(article)
abstracts.append(processed_abstract)
bar.next()
# sets url to next page in search
url = data['link'][-2]['@href']
# json file has 25 items per page, so go to the next page
item += 25
bar.finish()
# unreadable papers
print(f'Unreadable papers: {unreadable}')
# classifies and stores metadata
if abstracts:
self._store(articles, abstracts)
print()
else:
print('No abstracts to classify.\n')
return
# prints classifier metrics
for classifier in self._classifiers:
classifier.print_metrics()
classifier.reset_metrics()
# prints general tag metrics
if self._save:
print(f'Total articles analyzed: {self._gen_total}.')
print(
f'Stored {self._gen_new} new abstracts to \'{self._gen_tag}\'.'
)
print()
self._gen_new = 0
self._gen_total = 0
def scrape(self, subject='', keyword=''):
"""
Scrapes metadata of Springer Nature articles returned by subject and
keyword query, processes abstracts, and stores relevant articles
:param subject: subject constraint query, if empty does not include subject
constraint to query
:param keyword: keyword constraint query, if empty does not include keyword
constraint to query
"""
# prints subject and query made
subject_print = subject if subject else 'None'
keyword_print = keyword if keyword else 'None'
print(
f'Collection: {self._collection.database.name}.{self._collection.name}. Database: Springer Nature. Subject: {subject_print}, Keyword: {keyword_print}.'
)
articles = []
abstracts = []
unreadable = 0
no_doi = 0
item = 0
total = 100
# progress bar
bar = ChargingBar('Getting metadata:',
max=total,
suffix='%(index)d of %(max)d - %(elapsed_td)s')
while item < total:
# builds url and queries API
url = self._url_builder(item, subject, keyword)
response = requests.get(url)
if response.ok:
data = json.loads(response.content)
records = data['records']
# updates total to total number of papers in query
if item == 0:
total = int(data['result'][0]['total'])
bar.max = total
# gets metadata
for record in records:
# ignore abstract if doi is null
doi = record.get('doi')
if not doi:
no_doi += 1
bar.next()
continue
# store abstract text for use by mat2vec below
abstract = record.get('abstract')
# continues if paper does not have abstract
if not abstract:
unreadable += 1
bar.next()
continue
# segments abstract by sentence
doc = self.nlp(abstract)
sentences = []
is_unreadable = False
# processes sentence text using mat2vec processor
for sent in doc.sents:
try:
tokens, materials = self.processor.process(
sent.text)
except OverflowError:
is_unreadable = True
break
processed_sent = ' '.join(tokens)
sentences.append(processed_sent)
# if processor (from above) throws an error, skip the paper
if is_unreadable:
unreadable += 1
bar.next()
continue
processed_abstract = '\n'.join(sentences)
# create new document and store new article document if not in collection
article = {
'doi':
record.get('doi'),
'uid':
None,
'title':
record.get('title'),
'abstract':
abstract,
'url':
self._get_url(record.get('url')),
'creators':
self._get_creators(record.get('creators')),
'publication_name':
record.get('publicationName'),
'issn':
record.get('issn'),
'eissn':
record.get('eIssn'),
'publication_date':
self._get_date(record.get('publicationDate')),
'database':
'springer',
'processed_abstract':
processed_abstract
}
articles.append(article)
abstracts.append(processed_abstract)
bar.next()
# classify abstracts if 20000 have been stored
if len(abstracts) == 20000:
self._store(articles, abstracts)
articles = []
abstracts = []
# 100 items per page, so go to next page
item += 100
bar.finish()
# unreadable papers
print(f'No DOI: {no_doi}')
print(f'Unreadable papers: {unreadable}')
# classifies and stores metadata
if abstracts:
self._store(articles, abstracts)
print()
else:
print('No abstracts to classify.\n')
return
# prints classifier metrics
for classifier in self._classifiers:
classifier.print_metrics()
classifier.reset_metrics()
# prints general tag metrics
if self._save:
print(f'Total articles analyzed: {self._gen_total}.')
print(
f'Stored {self._gen_new} new abstracts to \'{self._gen_tag}\'.'
)
print()
self._gen_new = 0
self._gen_total = 0
def scrape(self, query):
"""
Scrapes metadata of Elsevier (ScienceDirect) articles returned
by query, processes abstracts, and stores relevant articles
:param query: Elsevier database query
"""
print(
f'Collection: {self._collection.database.name}.{self._collection.name}. Database: Elsevier. Query: {query}.'
)
# creates search url
url = f'https://api.elsevier.com/content/search/sciencedirect?query={query}&apiKey={ELSEVIER_API_KEY}&httpAccept=application%2Fjson'
# gets dois
dois = []
item = 0
total = 5000
# progress bar
bar = ChargingBar('Getting DOIs:',
max=total,
suffix='%(index)d of %(max)d - %(elapsed_td)s')
while item < total:
response = requests.get(url)
if response.ok:
data = json.loads(response.content)['search-results']
# updates total to total number of papers in query
if item == 0:
total = min(5000, int(data['opensearch:totalResults']))
bar.max = total
# stores dois
for entry in data['entry']:
doi = entry.get('prism:doi')
if doi:
dois.append(current_doi)
bar.next()
# sets url to next page in search
url = data['link'][-2]['@href']
# json file has 25 items per page, so go to the next page
item += 25
bar.finish()
# metadata
articles = []
abstracts = []
unreadable = 0
if not dois:
print('No abstracts to classify.\n')
return
# progress bar
bar = ChargingBar('Getting metadata:',
max=len(dois),
suffix='%(index)d of %(max)d - %(elapsed_td)s')
for doi in dois:
url = f'https://api.elsevier.com/content/article/doi/{doi}?apiKey={ELSEVIER_API_KEY}&httpAccept=application%2Fjson'
response = requests.get(url)
if response.ok:
try:
data = json.loads(
response.content
)['full-text-retrieval-response']['coredata']
except json.decoder.JSONDecodeError:
unreadable += 1
bar.next()
continue
# store abstract text for use by mat2vec below
abstract = data.get('dc:description')
# continues if paper does not have abstract
if not abstract:
unreadable += 1
bar.next()
continue
# segments abstract by sentence
doc = self.nlp(abstract)
sentences = []
is_unreadable = False
# processes sentence text using processor from mat2vec
for sent in doc.sents:
try:
tokens, materials = self.processor.process(sent.text)
except OverflowError:
is_unreadable = True
break
processed_sent = ' '.join(tokens)
sentences.append(processed_sent)
# if processor (from above) throws an error, skip the paper
if is_unreadable:
bar.next()
unreadable += 1
continue
processed_abstract = '\n'.join(sentences)
article = {
'doi': doi,
'uid': None,
'title': data.get('dc:title'),
'abstract': abstract,
'url': data.get('prism:url'),
'creators': self._get_creators(data.get('dc:creator')),
'publication_name': data.get('prism:publicationName'),
'issn': data.get('prism:issn'),
'publication_date':
self._get_date(data.get('prism:coverDate')),
'database': 'elsevier',
'processed_abstract': processed_abstract,
}
articles.append(article)
abstracts.append(processed_abstract)
bar.next()
bar.finish()
# unreadable papers
print(f'Unreadable papers: {unreadable}')
# classifies and stores metadata
if abstracts:
self._store(articles, abstracts)
print()
else:
print('No abstracts to classify.\n')
return
# prints classifier metrics
for classifier in self._classifiers:
classifier.print_metrics()
classifier.reset_metrics()
# prints general tag metrics
if self._save:
print(f'Total articles analyzed: {self._gen_total}.')
print(
f'Stored {self._gen_new} new abstracts to \'{self._gen_tag}\'.'
)
print()
self._gen_new = 0
self._gen_total = 0
