How to set the categorical slot value with the intent?

Rasa Open Source
1

I have the following categorical slot defined.

entities:
- login_id_type
slots:
  login_id_type:
    type: categorical
    influence_conversation: true
    values:
    - phone_verify
    - login_code
    - email

And set stories for each category as:

- story: how_to_login - phone_verify
  steps:
  - intent: how_to_login
  - action: utter_ask_login_id_type
  - intent: inform
  - slot_was_set:
    - login_id_type: phone_verify
  - action: utter_ans_login_id_type_phone_verify
- story: how_to_login - login_code
  steps:
  - intent: how_to_login
  - action: utter_ask_login_id_type
  - intent: inform
  - slot_was_set:
    - login_id_type: login_code
  - action: utter_ans_login_id_type_login_code
- story: how_to_login - email
  steps:
  - intent: how_to_login
  - action: utter_ask_login_id_type
  - intent: inform
  - slot_was_set:
    - login_id_type: email
  - action: utter_ans_login_id_type_email

But I’m not sure it’s the right way to set slot value as below:

- intent: inform
  examples: |
    - [AA]("entity": "login_id_type", "value": "login_code")
    - [AAA]("entity": "login_id_type", "value": "login_code")
    - [BB]("entity": "login_id_type", "value": "phone_verify")
    - [BBB]("entity": "login_id_type", "value": "phone_verify")
    - [CC]("entity": "login_id_type", "value": "email")
    - [CCC]("entity": "login_id_type", "value": "email")

I can train the model without errors. But it’s wrong to run the model

2020-11-26 19:48:39 DEBUG    rasa.core.processor  - Received user message 'BB' with intent '{'id': 2632104468417020721, 'name': 'inform', 'confidence': 0.9999431371688843}' and entities '[{'entity': '"entity"', 'start': 0, 'end': 2, 'confidence_entity': 0.7480142574109602, 'value': ' "login_id_type", "value": "phone_verify"', 'extractor': 'CRFEntityExtractor', 'processors': ['EntitySynonymMapper']}, {'entity': '"entity"', 'start': 0, 'end': 2, 'confidence_entity': 0.9065237045288086, 'value': 'BB', 'extractor': 'DIETClassifier'}]'
/home/sz/py36-bert110-grakn181-rasa203/lib/python3.7/site-packages/rasa/shared/utils/io.py:93: UserWarning: Interpreter parsed an entity '"entity"' which is not defined in the domain. Please make sure all entities are listed in the domain.
  More info at https://rasa.com/docs/rasa/domain
2020-11-26 19:48:39 DEBUG    rasa.core.processor  - Current slot values:
        feedback_message: None
        feedback_value: None
        login_id_type: None
        requested_slot: None
2020-11-26 19:48:39 DEBUG    rasa.core.processor  - Logged UserUtterance - tracker now has 8 events.
2020-11-26 19:48:39 DEBUG    rasa.core.policies.memoization  - Current tracker state [{}, {'user': {'intent': 'how_to_login'}, 'prev_action': {'action_name': 'action_listen'}}, {'user': {'intent': 'how_to_login'}, 'prev_action': {'action_name': 'utter_ask_login_id_type'}}, {'user': {'intent': 'inform'}, 'prev_action': {'action_name': 'action_listen'}}]
2020-11-26 19:48:39 DEBUG    rasa.core.policies.memoization  - Launch DeLorean...
2020-11-26 19:48:39 DEBUG    rasa.core.policies.memoization  - Current tracker state [{}, {'user': {'intent': 'inform'}, 'prev_action': {'action_name': 'action_listen'}}]
2020-11-26 19:48:39 DEBUG    rasa.core.policies.memoization  - There is no memorised next action
2020-11-26 19:48:39 DEBUG    rasa.core.policies.rule_policy  - Current tracker state: [{}, {'user': {'intent': 'how_to_login'}, 'prev_action': {'action_name': 'action_listen'}}, {'user': {'intent': 'how_to_login'}, 'prev_action': {'action_name': 'utter_ask_login_id_type'}}, {'user': {'intent': 'inform'}, 'prev_action': {'action_name': 'action_listen'}}]
2020-11-26 19:48:39 DEBUG    rasa.core.policies.rule_policy  - There is no applicable rule.
2020-11-26 19:48:39 DEBUG    rasa.core.policies.ensemble  - Predicted next action using policy_0_TEDPolicy
2020-11-26 19:48:39 DEBUG    rasa.core.processor  - Predicted next action 'utter_ans_login_id_type_email' with confidence 0.47.
2
2

I’d suggest using buttons and including the entities in the payload. For this you’ll set the payload equal to /inform{“login_id_type”: ”phone_verify”} for example. This will bypass the NLU pipeline and deliver the intent and entities directly to the dialog manager.

3

@kearnsw Due to the limitation of some channel, I can’t use buttons here. Without buttions, what should I do? Use non-categorial slot + synonym + action to get the job done?

4

Based on many many tries, synonym doesn’t work here. If anyone can give an example how to use categorial slot to do Contextual Conversations, that would be great.

5

Is this via SMS? It looks like you are offering options A,B, and C. You should implement this as a form which will provide validation and in the case of an incorrect value it’ll reprompt the user. Additionally, it’ll allow you to collapse your stories into a single story. For this you’ll need to use the from_entity() function.


forms:
  login_form:
    login_id_type:
      - type: from_entity
        entity: selection

I’d suggest using a regular expression match for these instead of going for an ML solution.


nlu:
- regex: selection
  examples: |
    - \bA\b
    - \bB\b
    - \bC\b
6

Thanks for your suggestion. A/B/C are not the only options. Acutally, there’re three categories indeed. There are just many synonyms for each category.

Form is good for collecting user information, I guess it’s a little heavy here. I prefer a Contextual Conversation. There should be a easy way to do it. Since the Rasa doc shows an exmple of Contextual Conversation with bool slots.

I guess, what I need is just an example of Contextual Conversation with categorial slots. Do you have any experience about it?

Thanks, again

7

It seems that no easy way to set categorical slot value with intent directly. But we can by pass the issue with synonym + categorical slot + contextual story.

8

Can you provide an example of your solution?

9

If you’re trying to set a categorical slot using an intent and have a single slot you are trying to fill, then the from_trigger_intent mapping should do the trick.


forms:
  login_form:
    login_id_type:
    - type: from_trigger_intent
      value: phone_verify
      intent: inform_login_id_phone_verify

However, I would still recommend a form with the from_intent method as you are starting with the same intent utter_ask_login_id_type for each context and incorporating this prompt into the form will provide better opportunity for error handling. Not to mention this is the exact syntax used for forms.

1 Like