Which is the best entity extractor for random names

Rasa Open Source
1

Hi,

I am trying to match entities that can contain any character [A-Za-z0-9_.,-"\s()’] etc. Can someone please tell me the best entity extractor to use

Examples:

  • Universal Plug N’ Play Event
  • Point-to-Point Tunnelling Protocol,ms-content-repl-srv
  • DirectInternet,VPN,ServiceLink,PrivateWAN,PrivateVPN
  • US-0402_ABC Airport Store
  • UT_St. Louis Sales
  • Hi-Tech Hley (NSW),Hy-Tech Bangalore (NSW)
2

There’s a blog post on this here

3

@stephens i had gone through the post. The question i had was would DIET suffice or would i need to include ner_crf for custom and random entities

4

I think DIET is fine. I’m assuming you can’t come up with a reasonable list of what the user could enter in this case. I think you should try to prompt the user for the value in such a way that you can use from_text to extract the entity?

1 Like
5

Yes. Thats is an option. But in our case the user enters the entire command which first reaches our backend and then to the nlp server. So that is actually not a useful option for us In most cases i am seeing that it is failing as some strings are not in training data and model is not able to generalize