KnowGL Knowledge Extractor

The knowgl-large model is trained by combining Wikidata with an extended version of the training data in the REBEL dataset. Given a sentence, KnowGL generates triple(s) in the following format:

[(subject mention # subject label # subject type) | relation label | (object mention # object label # object type)]

If there are more than one triples generated, they are separated by $ in the output. The model achieves state-of-the-art results for relation extraction on the REBEL dataset. The generated labels (for the subject, relation, and object) and their types can be directly mapped to Wikidata IDs associated with them.

This KnowledgeExtractor does not use any entity/relation pre-defined.

• Paper Rossiello et al. (AAAI 2023)
• Paper Mihindukulasooriya et al. (ISWC 2022)
• Original Model

Bases: KnowledgeExtractor

Instantiate the KnowGL Knowledge Extractor

load_models()

Load KnowGL model

parse_result(result, doc, encodings)

Parse the text result into a list of triples

Parameters:

Name	Type	Description	Default
result	str	Text generate by the KnowGL model	required
doc	Doc	Spacy doc	required
encodings	Encoding	Encodings result of the tokenization	required

Returns:

Type	Description
List[Tuple[Span, RelationSpan, Span]]	List of triples (subject, relation, object)

predict(docs, batch_size=None)

Extract triples from docs

Parameters:

Name	Type	Description	Default
docs	Iterator[Doc]	Spacy Docs to process	required
batch_size	Optional[Union[int, None]]	Batch size for processing	None

Returns:

Type	Description
List[List[Tuple[Span, RelationSpan, Span]]]	Triples (subject, relation, object) extracted for each document