"""
This module needs extra-dependencies not installed as core dependencies of medkit.
To install them, use `pip install medkit-lib[edsnlp]`.
"""
__all__ = ["EDSNLPDateMatcher"]
from typing import Iterator, List, Optional
import spacy
from medkit.core.text.operation import NEROperation
from medkit.core.text import Segment, Entity
from medkit.core.text import span_utils
from medkit.text.spacy.edsnlp import build_date_attribute, build_duration_attribute
[docs]class EDSNLPDateMatcher(NEROperation):
"""
Date matcher based on `EDS-NPL's dates pipeline
<https://aphp.github.io/edsnlp/latest/pipelines/misc/dates/>`_ which itself
relies on regular expressions. Note that this operation is designed to run
on french documents.
Absolute dates (ex: "23/08/2021"), relatives dates (ex: "la semaine
dernière") and durations (ex: "pendant quatre jours") will be matched.
For each date that is found, an entity will be created with an attribute
attached to it containing normalized values of the date components. The
attribute label will be either "date" or "duration", and the class of the
attribute will be either class :class:`~medkit.text.ner.DateAttribute`,
:class:`~medkit.text.ner.RelativeDateAttribute` or
:class:`~medkit.text.ner.DurationAttribute`.
"""
def __init__(
self,
output_label: str = "date",
attrs_to_copy: Optional[List[str]] = None,
uid: Optional[str] = None,
):
"""
Parameters
----------
output_label:
Label to use for date entities created (the label of the
attributes will always be "date" or "duration")
attrs_to_copy:
Labels of the attributes that should be copied from the input segment
to the created date entity. Useful for propagating context attributes
(negation, antecedent, etc).
uid:
Identifier of the matcher
"""
super().__init__(
output_label=output_label, attrs_to_copy=attrs_to_copy, uid=uid
)
if attrs_to_copy is None:
attrs_to_copy = []
self.output_label = output_label
self.attrs_to_copy = attrs_to_copy
self._edsnlp = spacy.blank("eds")
self._edsnlp.add_pipe("eds.dates")
[docs] def run(self, segments: List[Segment]) -> List[Entity]:
"""Find and return date entities for all `segments`
Parameters
----------
segments:
List of segments into which to look for date mentions
Returns
-------
entities: List[Entity]
Date entities found in `segments`, with
:class:`~medkit.text.ner.DateAttribute`,
:class:`~medkit.text.ner.RelativeDateAttribute` or
:class:`~medkit.text.ner.DurationAttribute` attributes.
"""
spacy_docs = self._edsnlp.pipe(s.text for s in segments)
return [
date_entity
for segment, spacy_doc in zip(segments, spacy_docs)
for date_entity in self._find_dates_in_segment(segment, spacy_doc)
]
def _find_dates_in_segment(self, segment, spacy_doc) -> Iterator[Entity]:
for spacy_span in spacy_doc.spans["dates"]:
yield self._build_entity(segment, spacy_span, is_duration=False)
for spacy_span in spacy_doc.spans["durations"]:
yield self._build_entity(segment, spacy_span, is_duration=True)
def _build_entity(self, segment, spacy_span, is_duration) -> Entity:
# convert span span start/end to medkit spans relative to doc
text, spans = span_utils.extract(
segment.text,
segment.spans,
[(spacy_span.start_char, spacy_span.end_char)],
)
# create attribute storing normalized date or duration values
attr = (
build_duration_attribute(spacy_span=spacy_span, spacy_label="duration")
if is_duration
else build_date_attribute(spacy_span=spacy_span, spacy_label="date")
)
# create entity
entity = Entity(label=self.output_label, spans=spans, text=text, attrs=[attr])
# handle provenance
if self._prov_tracer is not None:
self._prov_tracer.add_prov(
entity, self.description, source_data_items=[segment]
)
self._prov_tracer.add_prov(
attr, self.description, source_data_items=[segment]
)
# copy attrs from source segment to date entity
for label in self.attrs_to_copy:
for attr in segment.attrs.get(label=label):
copied_attr = attr.copy()
entity.attrs.add(copied_attr)
# handle provenance
if self._prov_tracer is not None:
self._prov_tracer.add_prov(copied_attr, self.description, [attr])
return entity
