Implement triple extractor

[albertus-andito]

AC:

• extractor should extract Subject, Predicate, Object from sentences

[albertus-andito]

Possible things to do:

1. Do coreference resolution
2. Extract SPO triples using OpenIE
3. Map SPO triples with DBpedia, using:
   • DBpedia Spotlight for S/O named entities
   • do some relation matching

[albertus-andito]

Implemented so far:

• Neuralcoref for coreference resolution
• OpenIE for SPO triples extraction
• DBpedia Spotlight for matching Subject/Object to DBpedia entity resources.

To do next:

• Match relations
• Better quality of triples

[albertus-andito]

Implemented so far:

1. TripleExtractors: abstract class TripleExtractor, implementation StanfordExtractor (Stanford OpenIE) and IITExtractor (IIT OpenIE). New implementations can be added.
2. TripleProducer:
   • Neuralcoref for coreference resolution
   • Triple extraction using TripleExtractors
   • Remove stopwords from Subject and Object if extraction_scope is not 'all'
   • Filter in subjects and objects according to extraction_scope
   • DBpedia Spotlight for matching Subject/Object to DBpedia entity resources
   • Link relations to DBpedia Ontology using Falcon
   • Lemmatise relations
   • Convert relations to DBpedia format/ontology (simple/without falcon)

[albertus-andito]

Basic working implementation is merged to main branch. see PR #22