Where is machine learning applied if I have to provide all dialogues in stories

Rasa Open Source
1

If we need to provide all possible scenarios in the stories.md, what is the purpose of the model?..where is machine learning applied?

Extracting entities, find the intents are done through the libraries…I am not able to understand how machine learning is helping here. Experts, please help.

@erohmensing ,@tmbo

1 Like
2

If we need to provide all possible scenarios in the stories.md, what is the purpose of the model?..where is machine learning applied?

You don’t need to provide all possible scenarios, if the KerasPolicy is part of your policy-ensemble (which is the default)

Extracting entities, find the intents are done through the libraries…I am not able to understand how machine learning is helping here. Experts, please help.

Whatever ‘are done through the libraries’ means, both task can and are usually done (again default in RASA) using machine learning. For entity extraction there’s the CRFEntity-Extractor used in RASA and for intent classification, there is the SklearnIntentClassifier commonly used in RASA pipelines.

CRFEntityExtractor

SKLearnIntentClassifier

1 Like
3

thanks @IgNoRaNt23. I am expecting some thing below,

Question1: If my stories.md have a intent and nlu.md as below

====================== ==========================================

story - ask office hours

  • intent_ask_office_hours

    -utter_office_hours.

================

In nlu.md,

  • intent: intent_ask_office_hours

    • can you provide office hours for tomorrow?
    • will the office open on next monday?
    • what is the office hours on weekends?

================

After training the model, if the user asks as question as " Is the office open today?" - bot should identify the intent, will the model handle this?..so far it is not working for me. Am i missing anything?

================= ==============================================

1 Like
4

you probably need more than 3 examples

5

Are you handling the datetime with an entity extractor? E.g. Duckling?

I’m seeing a normal utter-response, but you want to use a custom action for dynamic responses when handling time-dependent questions.

6

@IgNoRaNt23 I added around 10 examples, but still, bot is not able to understand if the question format is not added in the nlu.md.

This is where I am confused. If the bot can predict only the intent whatever we mention in the stories.md , where is machine learning helps here?..

2 Likes
7

Yes, I use duckling, but to explain, I gave this simple use-case. Extracting the date and time is good with duckling extractor, but for the response should we need to write custom actions?

For example,

GROUP1 QUESTIONS:

  • can you provide office hours for tomorrow?
  • will the office open on next monday?
  • what is the office hours on weekends?

GROUP2 QUESTIONS

  • Is the office open on 7 AM?
  • Is the office open quarter past noon?
  • do you close the office early ,before 4PM?

For the questions above, the response should be

GROUP1 QUESTIONS - Office hours on “requested_day” is 9AM to 4PM.

GROUP2 QUESTIONS - Yes, the office is open at " requested time" on “requested_day”.

For the above two groups, should we need to write two different actions?.. Shouldn’t the bot understand this difference and respond appropriately?

@IgNoRaNt23 @Krogsager @Juste

1 Like