Feature request: Custom NER model with spacy

[kapilkathuria]

Rasa NLU version: 0.13.4

As of now for creating a custom NER model, only option is CRF. I have multiple use cases where-in I need to extract data from documents / Pdfs. I haven't seen good results so far using CRF. I wanted to try spacy custom model.

As of now, I am creating spacy model via python script provided by spacy but I thought it will be good to get option out of box in RASA NLU to create custom spacy model.

[akelad]

I wouldn't really recommend using Rasa NLU to extract entities from documents/PDFs, it's for extracting entities from shorter sentences, mainly for chatbots.
I think using your own python script for this is the correct approach in this case

[kapilkathuria]

@akelad thanks for your note. Just in case you aware of any starting point (python script or any other NLU library) for, please share.

[akelad]

i'm not really sure, the spacy documentation is maybe a good starting point. I'll close this issue for now

[prashant334]

dn't really recommend using Rasa NLU to extract entities from documents/PDFs, it's for extracting entities from shorter sentences, main

@akelad Even for short sentences CRF BASED "ner_crf" is not giving accurate results. I have trained model to extract source and destination from short sentences like
show me the trains from Delhi to pune where DELHI = SOURCE and PUNE=DESTINATION

please find trained data in below attachment.

data.txt

BELOW IS MY CONFIG FILE.

language: "en"

pipeline:

• name: "SpacyNLP"
• name: "SpacyTokenizer"
  #- name: "SpacyFeaturizer"
  #- name: "RegexFeaturizer"
  #- name: "CRFEntityExtractor"
  #- name: "EntitySynonymMapper"
  #- name: "SklearnIntentClassifier"
• name: "ner_crf"
• name: "ner_spacy"

--