Padatious Intent Parser

Is there any document/paper/description on how the padatious intent parser is working under the hood? I could not find anything except the official documentation which doesn’t cover this topic very much.

I’m interested in how the intent and entity classification works.

I can see that it is based on the FANN library and uses simple feedforward nets but it would be great to get some more details about that

1 Like

Hey @marco ,

Thanks for reaching out - this is a GREAT question. I, unfortunately, don’t have much information about this topic.

I’m fairly certain @KathyReid could point us in the right direction. Thoughts, Kathy?


1 Like

Funny, was talking exactly about this in mattermost, i can nearly copy paste an answer for you :stuck_out_tongue:

  • padaos ensures exact matches by using regex
  • each padatious intent is a neural network model that tells you if for a piece f text you should trigger it, with a probability
  • padatious engine chooses the highest scoring intent, or None if all probabilities are bellow a threshold

Specifically, it uses one-hot encoding via local vocabularies (and a few extras like an unknown ratio input) into a series of shallow feed forward networks to perform intent recognition and entity extraction.


Great question, and thanks for the ping @darren-mycroft.

Unfortunately there isn’t a lot of extra documentation on Padatious - although @Wolfgange might have some comments to add when he has a spare few minutes.

1 Like

Yeah, what @Jarbas_Ai said pretty much sums it up. If you have any specific questions, or want more details/clarification, feel free to ask me here or on Mattermost (@matthew-mycroft)


Thanks for the quick answers everyone!

1 Like

@marco No problem, happy to help!

@Jarbas_Ai, @KathyReid @Wolfgange - thanks for the help everyone!

1 Like