class AhoCorasickEntityAnalyzer(Analyzer):
ORIGIN = 'regex'
MIN_LENGTH = 100
cache = AutomatonCache()
def prepare(self):
self.disabled = not get_config('REGEX_ENTITIES', True)
if not self.disabled:
self.cache.generate()
self.collector = DocumentTagCollector(self.document, self.ORIGIN)
def on_text(self, text):
if self.cache.automaton.kind == EMPTY:
return
if text is None or len(text) <= self.MIN_LENGTH:
return
text = match_form(text)
if text is None or len(text) <= self.MIN_LENGTH:
return
text = text.encode('utf-8')
for match in self.cache.automaton.iter(text):
for (text, type) in match[1]:
self.collector.emit(text, type)
def finalize(self):
log.info('Aho Corasick extraced %s entities.', len(self.collector))
self.collector.save()
def analyze(self, document):
if not document.supports_nlp:
return
collector = DocumentTagCollector(document, self.ORIGIN)
try:
self.extract(collector, document)
finally:
collector.save()
def analyze(self, document):
if document.schema in self.IGNORED:
return
collector = DocumentTagCollector(document, self.ORIGIN)
try:
self.extract(collector, document)
finally:
collector.save()
def update(self):
"""Apply the outcome of the result to the document."""
if self.status == self.STATUS_SUCCESS:
self.document.status = Document.STATUS_SUCCESS
self.document.error_message = None
else:
self.document.status = Document.STATUS_FAIL
self.document.type = Document.TYPE_OTHER
self.document.error_message = self.error_message
self.document.foreign_id = stringify(self.id)
if self.checksum:
self.document.content_hash = self.checksum
self.document.uploader_id = self.role_id or self.document.uploader_id
self.document.file_size = self.size
self.document.file_name = self.file_name
self.document.title = stringify(self.title)
self.document.summary = stringify(self.summary)
self.document.author = stringify(self.author)
self.document.keywords = self.keywords
self.document.mime_type = stringify(self.mime_type)
self.document.encoding = self.encoding
self.document.languages = self.languages
self.document.headers = self.headers
self.document.pdf_version = self.pdf_hash
self.document.columns = self.columns.keys()
collector = DocumentTagCollector(self.document, 'ingestors')
for entity in self.entities:
collector.emit(entity, DocumentTag.TYPE_PERSON)
for email in self.emails:
collector.emit(email, DocumentTag.TYPE_EMAIL)
collector.save()
def extract(self, collector, document):
DocumentTagCollector(document, 'polyglot').save()
DocumentTagCollector(document, 'spacy').save()
try:
service = EntityExtractStub(self.channel)
texts = self.text_iterator(document)
entities = service.Extract(texts)
for entity in entities.entities:
type_ = self.TYPES.get(entity.type)
if type_ is None:
continue
collector.emit(entity.label, type_, weight=entity.weight)
log.info('Extracted %s entities.', len(collector))
except self.Error as e:
log.warning("gRPC [%s]: %s", e.code(), e.details())
def extract(self, collector, document):
DocumentTagCollector(document, 'polyglot').save()
DocumentTagCollector(document, 'spacy').save()
try:
service = EntityExtractStub(self.channel)
texts = self.text_iterator(document)
entities = service.Extract(texts)
for entity in entities.entities:
type_ = self.TYPES.get(entity.type)
if type_ is None:
continue
collector.emit(entity.label, type_, weight=entity.weight)
log.info('Extracted %s entities.', len(collector))
except self.Error as exc:
log.exception("gRPC Error: %s", self.SERVICE)
self.reset_channel()
class RegexAnalyzer(Analyzer):
REGEX = None
FLAG = None
def prepare(self):
# TODO: re-think this.
self.disabled = self.document.type == self.document.TYPE_TABULAR
self.collector = DocumentTagCollector(self.document, self.ORIGIN)
self.regex = re.compile(self.REGEX, self.FLAG)
def on_text(self, text):
if not self.disabled:
for mobj in self.regex.finditer(text):
self.on_match(mobj)
def finalize(self):
self.collector.save()
def extract(self, collector, document):
DocumentTagCollector(document, 'polyglot').save()
DocumentTagCollector(document, 'spacy').save()
try:
service = EntityExtractStub(self.channel)
texts = self.text_iterator(document)
entities = service.Extract(texts)
for entity in entities.entities:
if entity.type == ExtractedEntity.COUNTRY:
document.add_country(entity.label)
if entity.type == ExtractedEntity.LANGUAGE:
document.add_language(entity.label)
type_ = self.TYPES.get(entity.type)
# log.info('%s: %s', entity.label, type_)
if type_ is not None:
collector.emit(entity.label, type_, weight=entity.weight)
log.info('Extracted %s entities.', len(collector))
except self.Error as e:
log.warning("gRPC [%s]: %s", e.code(), e.details())
def update(self):
"""Apply the outcome of the result to the document."""
doc = self.document
if self.status == self.STATUS_SUCCESS:
doc.status = Document.STATUS_SUCCESS
doc.error_message = None
else:
doc.status = Document.STATUS_FAIL
doc.error_message = stringify(self.error_message)
schema = model['Document']
for flag, name in self.SCHEMATA:
if flag in self.flags:
schema = model[name]
doc.schema = schema.name
doc.foreign_id = safe_string(self.id)
doc.content_hash = self.checksum or doc.content_hash
doc.pdf_version = self.pdf_checksum
doc.title = self.title or doc.meta.get('title')
doc.file_name = self.file_name or doc.meta.get('file_name')
doc.file_size = self.size or doc.meta.get('file_size')
doc.summary = self.summary or doc.meta.get('summary')
doc.author = self.author or doc.meta.get('author')
doc.generator = self.generator or doc.meta.get('generator')
doc.mime_type = self.mime_type or doc.meta.get('mime_type')
doc.encoding = self.encoding or doc.meta.get('encoding')
doc.date = self.date or doc.meta.get('date')
doc.authored_at = self.created_at or doc.meta.get('authored_at')
doc.modified_at = self.modified_at or doc.meta.get('modified_at')
doc.published_at = self.published_at or doc.meta.get('published_at')
doc.message_id = self.message_id or doc.meta.get('message_id')
doc.in_reply_to = ensure_list(self.in_reply_to)
doc.columns = list(self.columns.keys())
doc.body_raw = self.body_html
doc.body_text = self.body_text
doc.headers = self.headers
for kw in self.keywords:
doc.add_keyword(safe_string(kw))
for lang in self.languages:
doc.add_language(safe_string(lang))
db.session.flush()
collector = DocumentTagCollector(doc, 'ingestors')
for entity in self.entities:
collector.emit(entity, DocumentTag.TYPE_PERSON)
for email in self.emails:
collector.emit(email, DocumentTag.TYPE_EMAIL)
collector.save()
def update(self):
"""Apply the outcome of the result to the document."""
doc = self.document
if self.status == self.STATUS_SUCCESS:
doc.status = Document.STATUS_SUCCESS
doc.error_message = None
else:
doc.status = Document.STATUS_FAIL
doc.error_message = stringify(self.error_message)
schema = model['Document']
for flag, name in self.SCHEMATA:
if flag in self.flags:
schema = model[name]
doc.schema = schema.name
doc.foreign_id = stringify(self.id)
doc.content_hash = self.checksum or doc.content_hash
doc.uploader_id = self.role_id or doc.uploader_id
doc.title = stringify(self.title) or doc.meta.get('title')
doc.file_name = stringify(self.file_name) or doc.meta.get('file_name')
doc.file_size = self.size or doc.meta.get('file_size')
doc.title = stringify(self.title) or doc.meta.get('title')
doc.summary = stringify(self.summary) or doc.meta.get('summary')
doc.author = stringify(self.author) or doc.meta.get('author')
doc.generator = stringify(self.generator) or doc.meta.get('generator')
doc.mime_type = stringify(self.mime_type) or doc.meta.get('mime_type')
doc.encoding = stringify(self.encoding) or doc.meta.get('encoding')
doc.date = self.date or doc.meta.get('date')
doc.authored_at = self.created_at or doc.meta.get('authored_at')
doc.modified_at = self.modified_at or doc.meta.get('modified_at')
doc.published_at = self.published_at or doc.meta.get('published_at')
for kw in self.keywords:
doc.add_keyword(kw)
for lang in self.languages:
doc.add_language(lang)
doc.headers = self.headers or doc.meta.get('headers')
doc.columns = self.columns.keys()
if len(self.pages):
doc.body_text = '\n\n'.join(self.pages)
collector = DocumentTagCollector(doc, 'ingestors')
for entity in self.entities:
collector.emit(entity, DocumentTag.TYPE_PERSON)
for email in self.emails:
collector.emit(email, DocumentTag.TYPE_EMAIL)
collector.save()
def analyze(self, document):
if document.type in [document.TYPE_TABULAR, document.TYPE_OTHER]:
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' or len(entity) == 1:
continue
label = ' '.join(entity)
if len(label) < 4 or len(label) > 200:
continue
collector.emit(label, self.TYPES[entity.tag])
except ValueError as ve:
log.info('NER value error: %r', ve)
except Exception as ex:
log.warning('NER failed: %r', ex)
finally:
log.info('Polyglot extracted %s entities.', len(collector))
collector.save()
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 analyze(self, document):
collector = DocumentTagCollector(document, self.ORIGIN)
for match in self.RE.finditer(document.text):
text = self.extract_match(document, match)
if text is not None:
collector.emit(text, self.TYPE)
collector.save()
class PolyglotEntityAnalyzer(Analyzer):
ORIGIN = 'polyglot'
MIN_LENGTH = 100
TYPES = {
'I-PER': DocumentTag.TYPE_PERSON,
'I-ORG': DocumentTag.TYPE_ORGANIZATION,
'I-LOC': DocumentTag.TYPE_LOCATION
}
def prepare(self):
self.disabled = self.document.type != self.document.TYPE_TEXT
self.collector = DocumentTagCollector(self.document, self.ORIGIN)
def on_text(self, text):
if text is None or len(text) <= self.MIN_LENGTH:
return
try:
hint_language_code = None
if len(self.document.languages) == 1:
hint_language_code = self.document.languages[0]
text = Text(text, hint_language_code=hint_language_code)
for entity in text.entities:
if entity.tag == 'I-LOC' or len(entity) == 1:
continue
label = ' '.join(entity)
if len(label) < 4 or len(label) > 200:
continue
self.collector.emit(label, self.TYPES.get(entity.tag))
except ValueError as ve:
log.info('NER value error: %r', ve)
except Exception as ex:
log.warning('NER failed: %r', ex)
def finalize(self):
log.info('Polyglot extracted %s entities.', len(self.collector))
self.collector.save()
def analyze(self, document):
text = match_form(document.text)
if text is None or len(text) <= self.MIN_LENGTH:
return
collector = DocumentTagCollector(document, self.ORIGIN)
self.cache.generate()
if self.cache.automaton.kind != EMPTY:
text = text.encode('utf-8')
for match in self.cache.automaton.iter(text):
for (text, tag) in match[1]:
collector.emit(text, tag)
log.info('Aho Corasick extraced %s entities.', len(collector))
collector.save()
def extract_document_tags(document):
if document.status != Document.STATUS_SUCCESS:
return
load_places()
log.info("Tagging [%s]: %s", document.id, document.name)
languages = list(document.languages)
if not len(languages):
languages = [settings.DEFAULT_LANGUAGE]
aggregator = EntityAggregator()
for text in document.texts:
aggregator.extract(text, languages)
DocumentTagCollector(document, 'polyglot').save()
DocumentTagCollector(document, 'spacy').save()
collector = DocumentTagCollector(document, 'ner')
for (label, category, weight) in aggregator.entities:
collector.emit(label, category, weight=weight)
log.info("Extracted tags: %s", len(collector))
collector.save()
db.session.add(document)
db.session.commit()
def analyze(self, document):
collector = DocumentTagCollector(document, 'corasick')
if self.automaton is None:
return
for text in document.texts:
if len(text) <= self.MIN_LENGTH:
continue
text = self.match_form(text)
if text is None:
continue
for match in self.automaton.iter(text):
for (match_text, tag) in match[1]:
collector.emit(match_text, tag)
if len(collector):
log.info('Aho Corasick extraced %s entities.', len(collector))
collector.save()
def analyze(self, document):
if document.schema in self.IGNORED:
return
collector = DocumentTagCollector(document, self.ORIGIN)
try:
languages = set(document.languages)
if len(self.languages):
languages = languages.intersection(self.languages)
if not len(languages):
languages = [settings.DEFAULT_LANGUAGE]
for text in document.texts:
if len(text) <= self.MIN_LENGTH:
continue
for label, tag in self.tag_text(text, languages):
# log.info("Entity [%s]: %s", document.id, label)
collector.emit(label, self.TYPES[tag])
except ValueError as ve:
log.warning('NER value error: %r', ve)
collector.save()
if len(collector):
log.info('Polyglot extracted %s entities.', len(collector))
def prepare(self):
self.disabled = not get_config('REGEX_ENTITIES', True)
if not self.disabled:
self.cache.generate()
self.collector = DocumentTagCollector(self.document, self.ORIGIN)
def prepare(self):
# TODO: re-think this.
self.disabled = self.document.type == self.document.TYPE_TABULAR
self.collector = DocumentTagCollector(self.document, self.ORIGIN)
self.regex = re.compile(self.REGEX, self.FLAG)
def prepare(self):
self.disabled = self.document.type != self.document.TYPE_TEXT
self.collector = DocumentTagCollector(self.document, self.ORIGIN)
