Subject-verb-object extraction as part of narrative network needs to be more accurate

[ThijsVroegh]

The current subject-verb-object extraction as part of narrative network needs to be more accurate.

• Account for verb combinations that are the 'action'. E.g.: "ik ben blij dat ik met hem ook een nieuwe vorm heb gevonden".
  subject: ik
  the verb should be : heb gevonden (combined group)
  object: een nieuwe vorm

• Account voor 'bijzinnen' ; split results for those. Currently, it appears that only for the final part after the comma, results are retrieved.

• E.g., "Vanuit een diep dal zijn we herrezen, we hebben onze draken in de ogen gekeken" -> results are
  we - gekeken - onze draken
  we - gekeken - de ogen

No results on the first part of the sentence (before the comma)

Consider using a slightly different approach to extract the SVO's, e.g. ,with:

create a function and apply to the text column of interest as shown:

import textacy
import spacy
nlp = spacy.load('en_core_web_sm')
import pandas as pd

def extract_SVOs(text):
doc = nlp(text)
tuples = textacy.extract.subject_verb_object_triples(doc)
return list(tuples)

df['new_column_with_SVOs'] = df['my_text_colum'].apply(extract_SVOs)

see https://stackoverflow.com/questions/50906510/how-to-build-a-subject-verb-object-extraction-model-in-python