#507 added task and labeling.

arekit/contrib/source/nerel/folding/fixed.py

@@ -4,7 +4,7 @@
 from arekit.common.folding.fixed import FixedFolding
 
 
-def create_fixed_folding(train_filenames, dev_filenames, test_filenames):
+def create_fixed_folding(train_filenames, dev_filenames, test_filenames, limit=None):
     """ Create fixed data-folding based on the predefined list of filenames,
         written in file.
     """
@@ -18,6 +18,10 @@ def create_fixed_folding(train_filenames, dev_filenames, test_filenames):
     for doc_id, filename in filenames_by_ids.items():
         ids_by_filenames[filename] = doc_id
 
+    train_filenames = train_filenames if limit is None else train_filenames[:limit]
+    test_filenames = test_filenames if limit is None else test_filenames[:limit]
+    dev_filenames = dev_filenames if limit is None else dev_filenames[:limit]
+
     fixed_folding = FixedFolding.from_parts({
         DataType.Train: [ids_by_filenames[filename] for filename in train_filenames],
         DataType.Test: [ids_by_filenames[filename] for filename in test_filenames],

arekit/contrib/source/nerel/io_utils.py

@@ -79,7 +79,7 @@ def map_doc_to_fold_type(version=DEFAULT_VERSION):
         return d2f
 
     @staticmethod
-    def read_dataset_split(version=DEFAULT_VERSION):
+    def read_dataset_split(version=DEFAULT_VERSION, docs_limit=None):
         f2d = {}
         for filename, split_type in NerelIOUtils.__iter_filename_and_splittype(version):
             if split_type not in f2d:

arekit/contrib/source/nerel/labels.py

@@ -0,0 +1,53 @@
+from arekit.common.labels.base import Label
+
+
+class OpinionBelongsTo(Label):
+    pass
+
+
+class OpinionRelatesTo(Label):
+    pass
+
+
+class NegEffectFrom(Label):
+    pass
+
+
+class NegStateFrom(Label):
+    pass
+
+
+class PosEffectFrom(Label):
+    pass
+
+
+class PosAuthorFrom(Label):
+    pass
+
+
+class NegAuthorFrom(Label):
+    pass
+
+
+class PosStateFrom(Label):
+    pass
+
+
+class NegativeTo(Label):
+    pass
+
+
+class PositiveTo(Label):
+    pass
+
+
+class AlternativeName(Label):
+    pass
+
+
+class StateBelongsTo(Label):
+    pass
+
+
+class OriginsFrom(Label):
+    pass

arekit/contrib/utils/pipelines/sources/nerel/doc_ops.py

@@ -0,0 +1,28 @@
+from arekit.common.experiment.api.ops_doc import DocumentOperations
+from arekit.contrib.source.nerel.io_utils import NerelVersions
+from arekit.contrib.source.nerel.reader import NerelDocReader
+
+
+class NERELDocOperation(DocumentOperations):
+    """ Document reader for the collection of the RuSentNE competition 2023.
+        For more details please follow the following repository:
+        github: https://github.com/dialogue-evaluation/RuSentNE-evaluation
+    """
+
+    def __init__(self, filename_by_id, version):
+        """ filename_ids: dict
+                Dictionary of {id: filename}, where
+                    - id: int
+                    - filename: str
+            version: SentiNerelVersions
+                Specify the appropriate version of teh SentiNEREL collection.
+        """
+        assert(isinstance(filename_by_id, dict))
+        assert(isinstance(version, NerelVersions))
+        super(NERELDocOperation, self).__init__()
+        self.__filename_by_id = filename_by_id
+        self.__version = version
+        self.__doc_reader = NerelDocReader(version)
+
+    def by_id(self, doc_id):
+        return self.__doc_reader.read_document(doc_id=doc_id, filename=self.__filename_by_id[doc_id])

arekit/contrib/utils/pipelines/sources/nerel/extract_text_relations.py

@@ -0,0 +1,58 @@
+from arekit.common.experiment.api.ops_doc import DocumentOperations
+from arekit.common.experiment.data_type import DataType
+from arekit.contrib.source.nerel.io_utils import NerelIOUtils, NerelVersions
+from arekit.contrib.utils.pipelines.sources.nerel.doc_ops import NERELDocOperation
+from arekit.contrib.utils.pipelines.sources.nerel.labels_fmt import NerelAnyLabelFormatter
+from arekit.contrib.utils.pipelines.text_opinion.annot.predefined import PredefinedTextOpinionAnnotator
+from arekit.contrib.utils.pipelines.text_opinion.extraction import text_opinion_extraction_pipeline
+from arekit.contrib.utils.pipelines.text_opinion.filters.distance_based import DistanceLimitedTextOpinionFilter
+from arekit.contrib.utils.pipelines.text_opinion.filters.entity_based import EntityBasedTextOpinionFilter
+
+
+def create_text_relation_extraction_pipeline(sentinerel_version,
+                                             text_parser,
+                                             label_formatter=NerelAnyLabelFormatter(),
+                                             terms_per_context=50,
+                                             doc_ops=None,
+                                             docs_limit=None,
+                                             entity_filter=None):
+    assert(isinstance(sentinerel_version, NerelVersions))
+    assert(isinstance(doc_ops, DocumentOperations) or doc_ops is None)
+
+    data_folding = None
+
+    if doc_ops is None:
+        # Default Initialization.
+        filenames_by_ids, data_folding = NerelIOUtils.read_dataset_split(version=sentinerel_version,
+                                                                         docs_limit=docs_limit)
+        doc_ops = NERELDocOperation(filename_by_id=filenames_by_ids,
+                                    version=sentinerel_version)
+
+    text_opinion_filters = [
+        EntityBasedTextOpinionFilter(entity_filter=entity_filter),
+        DistanceLimitedTextOpinionFilter(terms_per_context)
+    ]
+
+    predefined_annot = PredefinedTextOpinionAnnotator(doc_ops, label_formatter)
+
+    pipelines = {
+        DataType.Train: text_opinion_extraction_pipeline(text_parser=text_parser,
+                                                         get_doc_by_id_func=doc_ops.by_id,
+                                                         annotators=[predefined_annot],
+                                                         text_opinion_filters=text_opinion_filters),
+        DataType.Test: text_opinion_extraction_pipeline(text_parser=text_parser,
+                                                        get_doc_by_id_func=doc_ops.by_id,
+                                                        annotators=[predefined_annot],
+                                                        text_opinion_filters=text_opinion_filters),
+        DataType.Dev: text_opinion_extraction_pipeline(text_parser=text_parser,
+                                                       get_doc_by_id_func=doc_ops.by_id,
+                                                       annotators=[predefined_annot],
+                                                       text_opinion_filters=text_opinion_filters),
+    }
+
+    # In the case when we setup a default data-folding.
+    # There is a need to provide it, due to the needs in further.
+    if data_folding is not None:
+        return pipelines, data_folding
+
+    return pipelines

arekit/contrib/utils/pipelines/sources/nerel/labels_fmt.py

@@ -0,0 +1,25 @@
+from arekit.common.labels.str_fmt import StringLabelsFormatter
+from arekit.contrib.source.nerel import labels
+
+
+class NerelAnyLabelFormatter(StringLabelsFormatter):
+
+    def __init__(self):
+
+        stol = {
+            "OPINION_BELONGS_TO": labels.OpinionBelongsTo,
+            "OPINION_RELATES_TO": labels.OpinionRelatesTo,
+            "NEG_EFFECT_FROM": labels.NegEffectFrom,
+            "POS_EFFECT_FROM": labels.PosEffectFrom,
+            "NEG_STATE_FROM": labels.NegStateFrom,
+            "POS_STATE_FROM": labels.PosStateFrom,
+            "NEGATIVE_TO": labels.NegativeTo,
+            "POSITIVE_TO": labels.PositiveTo,
+            "STATE_BELONGS_TO": labels.StateBelongsTo,
+            "POS_AUTHOR_FROM": labels.PosAuthorFrom,
+            "NEG_AUTHOR_FROM": labels.NegAuthorFrom,
+            "ALTERNATIVE_NAME": labels.AlternativeName,
+            "ORIGINS_FROM": labels.OriginsFrom
+        }
+
+        super(NerelAnyLabelFormatter, self).__init__(stol=stol)