def _create_affiliated_company(aff_name, aff_rel_span, aff_url_span,
company_owner, context, emitter):
aff_href = aff_url_span[0].find(".//a") if len(aff_url_span) else None
aff_href = aff_href.get("href") if aff_href is not None else None
aff_rel = aff_rel_span[0].text if len(aff_rel_span) else None
company = emitter.make("Company")
company.add("name", aff_name)
company.add("website", aff_href)
company.make_id("meineabgeordneten.at", aff_name)
company_ownership = emitter.make("Ownership")
if aff_rel:
# Info is given that way: "GESELLSCHAFTER 50.00% (100.00...)"
match_percentage = re.search(r"(\d\d?\d?\.\d\d)", aff_rel)
aff_pct = match_percentage.group(1) if match_percentage else None
if aff_pct:
aff_type = collapse_spaces(aff_rel[:match_percentage.start()])
else:
aff_type = collapse_spaces(aff_rel)
print("AFFILIATE pct '{}' ownerType '{}'".format(aff_pct, aff_type))
company_ownership.add("percentage", aff_pct)
company_ownership.add("ownershipType", aff_type)
company_ownership.add("owner", company_owner.id)
company_ownership.add("asset", company.id)
company_ownership.make_id(company_owner.id, company.id)
emitter.emit(company)
emitter.emit(company_ownership)
context.log.info("CREATED COMPANY '" + aff_name +
"' and membership with id '" + company_ownership.id + "'")
def _extract_table_description(context, row, isWork):
description_sub_el = row.xpath(".//span[@class='bold']")
affiliated = None
if len(description_sub_el):
# Description has a main part and a sub part.
# The main part usually states the name of an organisation and
# the sub part the function of the person in that organization.
desc_main = description_sub_el[0]
# sometimes text is wrapped inside <a ... /> that links to organization website
desc_parent = desc_main.getparent()
href = None
if desc_parent.tag == "a":
href = desc_parent.get("href")
desc_parent.remove(desc_main)
desc_parent = desc_parent.getparent()
else:
desc_parent.remove(desc_main)
if isWork:
aff_div = desc_parent.xpath(
'.//div[contains(@class,"tochterfirmen")]')
if len(aff_div):
affiliated = aff_div[0]
desc_parent.remove(affiliated)
desc_sub = collapse_spaces(desc_parent.text_content())
description = collapse_spaces(desc_main.text_content())
context.log.info("PARSED MANDATE DESCRIPTION: {}, {}".format(
description, desc_sub))
return description, desc_sub, href, affiliated
def pdf_extract_page(self, file_path, temp_dir, page):
"""Extract the contents of a single PDF page, using OCR if need be."""
pagenum = page.get('number')
page_size = self._element_size(page)
is_ocr = False
texts = []
for text in page.findall('.//text'):
content = text.xpath('string()').strip()
content = collapse_spaces(content)
if len(content):
texts.append(content)
for image in page.findall('.//image'):
ratio = self._element_size(image) / page_size
if len(texts) < 2 or ratio > self.IMAGE_RATIO_FOR_OCR:
is_ocr = True
if is_ocr and self.manager.config.get('PDF_OCR_PAGES', True):
image_file = self.pdf_page_to_image(file_path, pagenum, temp_dir)
with open(image_file, 'rb') as fh:
text = self.extract_text_from_image(fh.read())
text = collapse_spaces(text)
if text is not None:
texts.append(text)
text = ' \n'.join(texts).strip()
self.result.emit_page(int(pagenum), text)
def parse_html(context, data, result):
context.log.info('Parse: %r', result.url)
title = result.html.findtext('.//title')
if title is not None and 'title' not in data:
data['title'] = title
seen = set()
for tag_query, attr_name in URL_TAGS:
for element in result.html.findall(tag_query):
attr = element.get(attr_name)
if attr is None:
continue
url = normalize_url(urljoin(result.url, attr))
if url is None or url in seen:
continue
seen.add(url)
tag = make_key((context.run_id, url))
if context.check_tag(tag):
continue
context.set_tag(tag, None)
data = {'url': url}
# Option to set the document title from the link text.
if context.get('link_title', False):
data['title'] = collapse_spaces(element.text_content())
elif element.get('title'):
data['title'] = collapse_spaces(element.get('title'))
context.emit(rule='fetch', data=data)
def crawl(context: Context):
xls_url = fetch_xls_url(context)
path = context.fetch_resource("source.xls", xls_url)
context.export_resource(path, XLS, title=context.SOURCE_TITLE)
xls = xlrd.open_workbook(path)
for sheet in xls.sheets():
headers = None
row0 = [h.convert_excel_cell(xls, c) for c in sheet.row(0)]
sections = [c for c in row0 if c is not None]
section = collapse_spaces(" / ".join(sections))
for r in range(1, sheet.nrows):
row = [h.convert_excel_cell(xls, c) for c in sheet.row(r)]
# after a header is found, read normal data:
if headers is not None:
data: Dict[str, List[str]] = {}
for header, cell in zip(headers, row):
if header is None:
continue
values = []
if isinstance(cell, datetime):
cell = cell.date()
for value in multi_split(stringify(cell), SPLITS):
if value is None:
continue
if value == "不明":
continue
if value is not None:
values.append(value)
data[header] = values
emit_row(context, sheet.name, section, data)
if not len(row) or row[0] is None:
continue
teaser = row[0].strip()
# the first column of the common headers:
if "告示日付" in teaser:
if headers is not None:
context.log.error("Found double header?", row=row)
# print("SHEET", sheet, row)
headers = []
for cell in row:
cell = collapse_spaces(cell)
header = context.lookup_value("columns", cell)
if header is None:
context.log.warning("Unknown column title",
column=cell,
sheet=sheet.name)
headers.append(header)
def parse_html(context, data, result):
context.log.info("Parse: %r", result.url)
for title in result.html.xpath(".//title/text()"):
if title is not None and "title" not in data:
data["title"] = title
include = context.params.get("include_paths")
if include is None:
roots = [result.html]
else:
roots = []
for path in include:
roots = roots + result.html.xpath(path)
seen = set()
for root in roots:
for tag_query, attr_name in URL_TAGS:
for element in root.xpath(tag_query):
attr = element.get(attr_name)
if attr is None:
continue
try:
url = urljoin(result.url, attr)
key = url
except Exception:
log.warning("Invalid URL: %r", attr)
continue
if url is None or key is None or key in seen:
continue
seen.add(key)
tag = make_key(context.run_id, key)
if context.check_tag(tag):
continue
context.set_tag(tag, None)
data["url"] = url
if data.get("title") is None:
# Option to set the document title from the link text.
if context.get("link_title", False):
data["title"] = collapse_spaces(element.text_content())
elif element.get("title"):
data["title"] = collapse_spaces(element.get("title"))
context.http.session.headers["Referer"] = url
context.emit(rule="fetch", data=data)
def parse_html(context, data, result):
context.log.info('Parse: %r', result.url)
title = result.html.findtext('.//title')
if title is not None and 'title' not in data:
data['title'] = title
include = context.params.get('include_paths')
if include is None:
roots = [result.html]
else:
roots = []
for path in include:
roots = roots + result.html.findall(path)
seen = set()
for root in roots:
for tag_query, attr_name in URL_TAGS:
for element in root.findall(tag_query):
attr = element.get(attr_name)
if attr is None:
continue
try:
url = normalize_url(urljoin(result.url, attr))
except Exception:
log.warning('Invalid URL: %r', attr)
continue
if url is None or url in seen:
continue
seen.add(url)
tag = make_key(context.run_id, url)
if context.check_tag(tag):
continue
context.set_tag(tag, None)
data = {'url': url}
# Option to set the document title from the link text.
if context.get('link_title', False):
data['title'] = collapse_spaces(element.text_content())
elif element.get('title'):
data['title'] = collapse_spaces(element.get('title'))
context.http.session.headers['Referer'] = url
if re.findall('publicId|firstResult', url):
print("----------------PRINTING URL----------------")
print(url)
context.emit(rule='fetch', data=data)
def cleanup_text(text):
if text is None:
return
match = CLEANUP.match(text)
if match is not None:
term = match.group('term')
return collapse_spaces(term)
def generate(text, keep_order=False):
text = stringify(text)
if text is None:
return
# this needs to happen before the replacements
text = text.lower()
text = remove_person_prefix(text)
# remove any text in brackets
text = BRACKETED.sub(WS, text)
# super hard-core string scrubbing
text = clean_strict(text)
text = replace_types(text)
if keep_order:
text = collapse_spaces(text)
else:
# final manicure, based on openrefine algo
parts = [p for p in text.split(WS) if len(p)]
text = WS.join(sorted(set(parts)))
if not len(text):
return None
return text
def parse_entry(context, node):
entity_name = node.findtext("./Entity")
if entity_name is not None:
entity = context.make("LegalEntity")
entity.add("name", entity_name.split("/"))
else:
entity = context.make("Person")
given_name = node.findtext("./GivenName")
entity.add("firstName", given_name)
last_name = node.findtext("./LastName")
entity.add("lastName", last_name)
entity.add("name", jointext(given_name, last_name))
entity.add("birthDate", node.findtext("./DateOfBirth"))
# ids are per country and entry type (individual/entity)
item = node.findtext("./Item")
schedule = node.findtext("./Schedule")
country = node.findtext("./Country")
if "/" in country:
country, _ = country.split("/")
entity.id = context.make_slug(country, schedule, item, strict=False)
entity.add("country", country)
sanction = h.make_sanction(context, entity)
sanction.add("program", schedule)
names = node.findtext("./Aliases")
if names is not None:
for name in names.split(", "):
name = collapse_spaces(name)
entity.add("alias", name)
entity.add("topics", "sanction")
context.emit(entity, target=True)
context.emit(sanction)
def generate(text, keep_order=False):
text = ensure_text(text)
if text is None:
return
# this needs to happen before the replacements
text = text.lower()
# try to remove personal prefix, such as Mr., Mrs.
text = PERSON_PREFIX.sub(WS, text)
# remove any text in brackets
text = BRACKETED.sub(WS, text)
# super hard-core string scrubbing
text = clean_strict(text)
text = COMPANY_TYPES.sub(company_type_replacer, text)
if keep_order:
text = collapse_spaces(text)
else:
# final manicure, based on openrefine algo
parts = [p for p in text.split(WS) if len(p)]
text = WS.join(sorted(set(parts)))
if not len(text):
return None
return text
def analyze(self, document):
if document.schema in self.IGNORED:
return
collector = DocumentTagCollector(document, self.ORIGIN)
text = document.text
if text is None or len(text) <= self.MIN_LENGTH:
return
try:
hint_language_code = None
if len(document.languages) == 1:
hint_language_code = document.languages[0]
text = Text(text, hint_language_code=hint_language_code)
for entity in text.entities:
if entity.tag == 'I-LOC':
continue
label = ' '.join(entity)
label = self.CLEAN.sub(' ', label)
label = collapse_spaces(label)
if ' ' not in label or len(label) < 4 or len(label) > 200:
continue
# log.info("Entity [Doc %s]: %s [%s]",
# document.id, label, entity.tag)
collector.emit(label, self.TYPES[entity.tag])
except ValueError as ve:
log.warning('NER value error: %r', ve)
except Exception as ex:
log.warning('NER failed: %r', ex)
finally:
collector.save()
log.info('Polyglot extracted %s entities.', len(collector))
def generate(text: Optional[str],
keep_order: bool = False,
keep_brackets: bool = False) -> Optional[str]:
text = stringify(text)
if text is None:
return None
# this needs to happen before the replacements
text = text.lower()
text = clean_entity_name(text)
if not keep_brackets:
# Remove any text in brackets
# This is meant to handle names of companies which include
# the jurisdiction, like: Turtle Management (Seychelles) Ltd.
text = BRACKETED.sub(WS, text)
# Super hard-core string scrubbing
text = clean_strict(text)
text = replace_types(text)
if keep_order:
text = collapse_spaces(text)
elif text is not None:
# final manicure, based on openrefine algo
parts = [p for p in text.split(WS) if len(p)]
text = WS.join(sorted(set(parts)))
if text is None or not len(text):
return None
return text
def index_form(texts):
"""Turn a set of strings into the appropriate form for indexing."""
results = []
total_len = 0
for text in texts:
# We don't want to store more than INDEX_MAX_LEN of text per doc
if total_len > INDEX_MAX_LEN:
# TODO: there might be nicer techniques for dealing with overly
# long text buffers?
results = list(set(results))
total_len = sum((len(t) for t in results))
if total_len > INDEX_MAX_LEN:
break
text = stringify(text)
if text is None:
continue
text = collapse_spaces(text)
total_len += len(text)
results.append(text)
# Make latinized text version
latin = latinize_text(text)
latin = stringify(latin)
if latin is None or latin == text:
continue
total_len += len(latin)
results.append(latin)
return results
def apply(self, record):
value = six.text_type(self.template)
for repl, ref in self.replacements.items():
ref_value = record.get(ref) or ''
ref_value = six.text_type(ref_value)
value = value.replace(repl, ref_value)
return collapse_spaces(value).strip()
def Extract(self, request, context):
text = request.text
if text is None or not len(text.strip()):
return
entity_count = 0
for language in request.languages:
if language not in LANGUAGES:
continue
try:
parsed = Text(text, hint_language_code=language)
for entity in parsed.entities:
label = ' '.join(entity)
label = CLEAN.sub(' ', label)
label = collapse_spaces(label)
if len(label) < 4 or len(label) > 200:
continue
if ' ' not in label:
continue
length = entity.end - entity.start
entity_count += 1
yield ExtractedEntity(label=label,
offset=entity.start,
length=length,
type=TYPES[entity.tag])
except Exception:
log.exception("Cannot extract. Language: %s", language)
log.info("Extract: extracted %s entities.", entity_count)
def _make_work_and_affiliates(person, context, description, description_sub,
startDate, endDate, emitter, org_website,
affiliates):
company_owner = emitter.make("Company")
company_owner.add("website", org_website)
membership = emitter.make("Membership")
membership.add("startDate", startDate)
membership.add("endDate", endDate)
_create_org_and_attach(emitter, context, company_owner, person,
description, membership, description_sub, startDate)
if affiliates is None:
return
for aff in affiliates.xpath('.//div[contains(@class,"tochterfirma")]'):
aff_name_span = aff.xpath(".//span[@class='tochterFirmaName']")
aff_url_span = aff.xpath('.//span[@class="tochterFirmaLink"]')
aff_rel_span = aff.xpath('.//span[@class="tochterFirmaBeziehung"]')
aff_name = collapse_spaces(
aff_name_span[0].text) if len(aff_name_span) else None
if not aff_name:
# An affiliated company without a name indicates a parsing error.
continue
_create_affiliated_company(aff_name, aff_rel_span, aff_url_span,
company_owner, context, emitter)
def emit_document(context, row, date):
context.http.reset()
url = context.params.get("url")
cells = row.findall("./td")
if not len(cells):
return
text = [c.text_content().strip() for c in cells]
_, num, category, name, _ = text
title = "%s (%s, %s)" % (name, category, date)
title = collapse_spaces(title)
link = row.find('.//a[@class="pdfLnk"]')
if link is None:
return
url = urljoin(url, link.get("href"))
context.emit(
data={
"url": url,
"title": title,
"foreign_id": url,
"countries": ["ch"],
"dates": [date],
"extension": "pdf",
"mime_type": "application/pdf",
})
def crawl(context: Context):
path = context.fetch_resource("source.html", context.dataset.data.url)
context.export_resource(path, HTML, title=context.SOURCE_TITLE)
with open(path, "r") as fh:
doc = html.parse(fh)
table = doc.find('//div[@class="sanctioned-table"]/table')
headers = None
for row in table.findall(".//tr"):
if headers is None:
headers = [slugify(el.text) for el in row.findall("./th")]
continue
cells = [collapse_spaces(el.text) for el in row.findall("./td")]
data = {hdr: c for hdr, c in zip(headers, cells)}
entity = context.make("Person")
entity.id = context.make_id(data["id"], data["ad-soyad-ata-adi"])
entity.add("name", data["ad-soyad-ata-adi"])
entity.add("idNumber", data["id"])
entity.add("birthDate", parse_date(data["dogum-tarixi"]))
entity.add("country", "az")
entity.add("topics", "sanction")
addr = h.make_address(context, full=data["malumat"])
h.apply_address(context, entity, addr)
sanction = h.make_sanction(context, entity)
context.emit(sanction)
context.emit(entity, target=True)
def crawl(context: Context):
path = context.fetch_resource("source.html", context.dataset.data.url)
context.export_resource(path, HTML, title=context.SOURCE_TITLE)
with open(path, "r", encoding="utf-8") as fh:
doc = html.fromstring(fh.read())
for table in doc.findall('.//div[@class="editor-content"]//table'):
headers = None
schema = None
for row in table.findall(".//tr"):
cells = [
collapse_spaces(c.text_content()) for c in row.findall("./td")
]
if headers is None:
headers = [slugify(c, sep="_") for c in cells]
continue
if len(cells) == 1:
schema = TYPES[cells[0]]
continue
row = dict(zip(headers, cells))
entity = context.make(schema)
name = row.pop("imie_i_nazwisko_nazwa_podmiotu")
entity.id = context.make_slug(name)
names = name.split("(")
entity.add("name", names[0])
for alias in names[1:]:
entity.add("alias", alias.split(")")[0])
notes = row.pop("uzasadnienie_wpisu_na_liste")
entity.add("notes", notes)
details = row.pop("dane_identyfikacyjne_osoby_podmiotu")
for (chop, prop) in CHOPSKA:
parts = details.rsplit(chop, 1)
details = parts[0]
if len(parts) > 1:
if prop == "address":
addr = h.make_address(context, full=parts[1])
h.apply_address(context, entity, addr)
else:
entity.add(prop, parts[1])
if len(details.strip()):
result = context.lookup("details", details)
if result is None:
context.log.warning("Unhandled details", details=details)
else:
for prop, value in result.props.items():
entity.add(prop, value)
sanction = h.make_sanction(context, entity)
provisions = row.pop("zastosowane_srodki_sankcyjne")
sanction.add("provisions", provisions)
start_date = row.pop("data_umieszczenia_na_liscie")
start_date = start_date.replace(" r.", "")
sanction.add("startDate", h.parse_date(start_date, ["%d.%m.%Y"]))
h.audit_data(row)
context.emit(entity, target=True)
context.emit(sanction)
def clean_name(self, text):
if text is None or len(text) > self.MAX_LENGTH:
return
text = clean_entity_name(text)
text = collapse_spaces(text)
if len(text) < self.MIN_LENGTH:
return
return text
def _convert_time_span(raw_time_span):
raw_time_span = collapse_spaces(raw_time_span)
if "seit" in raw_time_span.lower():
# "seit" (since) indicates that there is no end time
return _parse_single_date(raw_time_span), None
elif "-" in raw_time_span:
arr = raw_time_span.split("-")
return _parse_single_date(arr[0]), _parse_single_date(arr[1])
def clean_name(text):
if text is None or len(text) > NAME_MAX_LENGTH:
return
text = clean_entity_name(text)
text = collapse_spaces(text)
if text is None or len(text) <= NAME_MIN_LENGTH or ' ' not in text:
return
return text
def name(self):
if self._name is not None:
return self._name
names = (self.first_name, self.second_name, self.third_name,
self.father_name, self.last_name)
names = [n for n in names if n is not None]
if len(names):
names = ' '.join(names)
return collapse_spaces(names)
def clean_strict(text, boundary=WS):
"""Super-hardcore string scrubbing."""
# transliterate to ascii
text = ascii_text(text)
# replace punctuation and symbols
text = CHARACTERS_REMOVE_RE.sub('', text)
text = category_replace(text)
# pad out for company type replacements
text = ''.join((boundary, collapse_spaces(text), boundary))
return text
def clean_name(cls, text):
if text is None or len(text) > MAX_LENGTH:
return
match = CLEANUP.match(text)
if match is not None:
text = match.group('term')
text = collapse_spaces(text)
if not len(text) or len(text) < MIN_LENGTH:
return
return text
def clean_strict(text, boundary=WS):
"""Super-hardcore string scrubbing."""
# transliterate to ascii
text = ascii_text(text)
# replace punctuation and symbols
text = CHARACTERS_REMOVE_RE.sub('', text)
text = category_replace(text)
# pad out for company type replacements
text = ''.join((boundary, collapse_spaces(text), boundary))
return text
def crawl(context: Context):
path = context.fetch_resource("source.html", context.dataset.data.url)
context.export_resource(path, HTML, title=context.SOURCE_TITLE)
with open(path, "r", encoding="ISO-8859-1") as fh:
doc = html.parse(fh)
table = doc.find("//div[@id='viewcontainer']/table")
headers = None
for row in table.findall(".//tr"):
if headers is None:
headers = [
slugify(c.text_content(), "_") for c in row.findall("./th")
]
continue
cells = [
collapse_spaces(c.text_content()) for c in row.findall("./td")
]
cells = dict(zip(headers, cells))
cells.pop(None, None)
full_name = name = cells.pop("name")
registration_number = None
for splitter in REG_NRS:
if splitter in name:
name, registration_number = name.split(splitter, 1)
registration_number = registration_number.replace(")", "")
country = cells.pop("nationality")
country = country.replace("Non ADB Member Country", "")
country = country.replace("Rep. of", "")
entity = context.make("LegalEntity")
entity.id = context.make_id(full_name, country)
entity.add("name", name)
entity.add("alias", cells.pop("othername_logo"))
entity.add("topics", "debarment")
entity.add("country", country)
entity.add("registrationNumber", registration_number)
sanction = h.make_sanction(context, entity)
sanction.add("reason", cells.pop("grounds"))
sanction.add("program", cells.pop("sanction_type"))
date_range = cells.pop("effect_date_lapse_date", "")
if "|" in date_range:
start_date, end_date = date_range.split("|")
sanction.add("startDate", h.parse_date(start_date.strip(),
FORMATS))
sanction.add("endDate", h.parse_date(end_date.strip(), FORMATS))
address = h.make_address(context,
full=cells.pop("address"),
country=country)
h.apply_address(context, entity, address)
context.emit(entity, target=True)
context.emit(sanction)
def get_countries(context):
doc = context.fetch_html(COUNTRIES_URL)
path = ".//select[@id='arrestWarrantCountryId']//option"
options = []
for option in doc.findall(path):
code = stringify(option.get("value"))
if code is None:
continue
label = collapse_spaces(option.text_content())
options.append((code, label))
return list(sorted(options))
def parse_alias(entity, node):
names = node.findtext('./ALIAS_NAME')
if names is None:
return
for name in names.split('; '):
name = collapse_spaces(name)
if not len(name):
continue
alias = entity.create_alias(name=name)
alias.quality = QUALITY[node.findtext('./QUALITY')]
def text_chunks(texts, sep=" ", max_chunk=25000):
"""Pre-chew text snippets for NLP and pattern matching."""
for text in texts:
text = collapse_spaces(text)
if text is None or len(text) < 5:
continue
# Crudest text splitting code in documented human history.
# Most of the time, a single page of text is going to be
# 3000-4000 characters, so this really only kicks in if
# something weird is happening in the first place.
for idx in range(0, len(text), max_chunk):
yield text[idx : idx + max_chunk]
