Automatic Open Wikidata Information Extraction from Indonesian Text

Abstract Open Information Extraction can be used to aid NLP pipeline, especially if it is linked to a knowledge graph. This is the first attempt to perform OIE in Indonesian text by leveraging pre-existing massive knowledge graph such as Wikidata. The model were given "easier" dataset sample to classify and "harder" dataset sample to classify based on its predicate's RDFS label. It achieves 70.14% F1 on the easier one and 46.68% F1 on the harder one. We also experimented with a pipeline that extracts triple "from the open", but the model struggles with triple that conveys relationship with two geographical place.