annif/corpus/skos.py
"""Support for subjects loaded from a SKOS/RDF file"""
from __future__ import annotations
import collections
import os.path
import shutil
from typing import TYPE_CHECKING
import rdflib
import rdflib.util
from rdflib.namespace import OWL, RDF, RDFS, SKOS
import annif.util
from .types import Subject, SubjectCorpus
if TYPE_CHECKING:
from collections.abc import Iterator, Sequence
from rdflib.term import URIRef
def serialize_subjects_to_skos(subjects: Iterator, path: str) -> None:
"""Create a SKOS representation of the given subjects and serialize it
into a SKOS/Turtle file with the given path name."""
import joblib
graph = rdflib.Graph()
graph.namespace_manager.bind("skos", SKOS)
for subject in subjects:
graph.add((rdflib.URIRef(subject.uri), RDF.type, SKOS.Concept))
for lang, label in subject.labels.items():
graph.add(
(
rdflib.URIRef(subject.uri),
SKOS.prefLabel,
rdflib.Literal(label, lang),
)
)
graph.add(
(
rdflib.URIRef(subject.uri),
SKOS.notation,
rdflib.Literal(subject.notation),
)
)
graph.serialize(destination=path, format="turtle")
# also dump the graph in joblib format which is faster to load
annif.util.atomic_save(
graph, *os.path.split(path.replace(".ttl", ".dump.gz")), method=joblib.dump
)
class SubjectFileSKOS(SubjectCorpus):
"""A subject corpus that uses SKOS files"""
PREF_LABEL_PROPERTIES = (SKOS.prefLabel, RDFS.label)
_languages = None
def __init__(self, path: str) -> None:
self.path = path
if path.endswith(".dump.gz"):
import joblib
self.graph = joblib.load(path)
else:
self.graph = rdflib.Graph()
self.graph.parse(self.path, format=rdflib.util.guess_format(self.path))
@property
def languages(self) -> set[str]:
if self._languages is None:
self._languages = {
label.language
for concept in self.concepts
for label_type in self.PREF_LABEL_PROPERTIES
for label in self.graph.objects(concept, label_type)
if label.language is not None
}
return self._languages
def _concept_labels(self, concept: URIRef) -> dict[str, str]:
by_lang = self.get_concept_labels(concept, self.PREF_LABEL_PROPERTIES)
return {
lang: (
by_lang[lang][0]
if by_lang[lang] # correct lang
else (
by_lang[None][0]
if by_lang[None] # no language
else self.graph.namespace_manager.qname(concept)
)
)
for lang in self.languages
}
@property
def subjects(self) -> Iterator[Subject]:
for concept in self.concepts:
labels = self._concept_labels(concept)
notation = self.graph.value(concept, SKOS.notation, None, any=True)
if notation is not None:
notation = str(notation)
yield Subject(uri=str(concept), labels=labels, notation=notation)
@property
def concepts(self) -> Iterator[URIRef]:
for concept in self.graph.subjects(RDF.type, SKOS.Concept):
if (concept, OWL.deprecated, rdflib.Literal(True)) in self.graph:
continue
yield concept
def get_concept_labels(
self,
concept: URIRef,
label_types: Sequence[URIRef],
) -> collections.defaultdict[str | None, list[str]]:
"""return all the labels of the given concept with the given label
properties as a dict-like object where the keys are language codes
and the values are lists of labels in that language"""
labels_by_lang = collections.defaultdict(list)
for label_type in label_types:
for label in self.graph.objects(concept, label_type):
labels_by_lang[label.language].append(str(label))
return labels_by_lang
@staticmethod
def is_rdf_file(path: str) -> bool:
"""return True if the path looks like an RDF file that can be loaded
as SKOS"""
fmt = rdflib.util.guess_format(path)
return fmt is not None
def save_skos(self, path: str) -> None:
"""Save the contents of the subject vocabulary into a SKOS/Turtle
file with the given path name."""
if self.path.endswith(".ttl"):
# input is already in Turtle syntax, no need to reserialize
if not os.path.exists(path) or not os.path.samefile(self.path, path):
shutil.copyfile(self.path, path)
else:
# need to serialize into Turtle
self.graph.serialize(destination=path, format="turtle")
# also dump the graph in joblib format which is faster to load
import joblib
annif.util.atomic_save(
self.graph,
*os.path.split(path.replace(".ttl", ".dump.gz")),
method=joblib.dump,
)