Problem while differentiating stories

Rasa Open Source
1

Hi,

We’re having problems differentiating the stories. If the user enters the text “Habt ihr eine Kita”, the Intent “info” is recognized correctly and the entity “MISC” with the value “Kita” is recognized as well. However, the utterance (utter_info_kita) corresponding to the story is not predicted. All 3 utterances have nearly the same prediction value at round about 33%.

## inform about Krankenhaus
* info{"MISC":"Krankenhaus"}
  - slot{"MISC": "Krankenhaus"}
  - utter_info_krankenhaus
 
## inform about Kita
* info{"MISC":"Kita"}
  - slot{"MISC": "Kita"}
  - utter_info_kita
 
## inform about Seniorenheim
* info{"MISC":"Seniorenheim"}
  - slot{"MISC": "Seniorenheim"}
  - utter_info_seniorenheim

Here is the entry in the nlu.md file:

## intent:info
- Was ist ein [Krankenhaus](MISC)?
- Habt ihr eine [Kita](MISC)?
- Habt ihr ein [Seniorenheim](MISC)?
- [Krankenhaus](MISC)
- [Kita](MISC)
- [Seniorenheim](MISC)
- Gib mir Informationen zum [Krankenhaus](MISC).
- Gib mir Informationen zur [Kita](MISC).

There are no errors during training. Does anybody know a solution for this problem?

2

If your slot MISC is not categorical Rasa doesn’t take into account its value, only if it has one or not. If that’s your case you can change MISC to a categorical slot.

Another option is to replace your “utter_info” actions with a single custom action that selects a response in accordance to the slot value.

2 Likes
3

You’re absolutely right. After changing the slot type to categegorical it works for us. Thanks a lot!