DialogFlow CX - German - @sys.person takes wrong names

[tschiffler]

I use DialogFlow CX with German als Language
There I trained a intent that tries to extract the name from a utterance

Here I have an entity that uses the sys.person system entity

Today I found a strange Problem that can be reproduced by the test
I enter "nein danke ciao" as utterance (that means something like 'no thanks bye'). The intent is catched with a confidence of 1 with the match on the system entity

"Id": "XXXX",
"DisplayName": "INTENT_NAME",
"Score": 1,
"Parameters": {
 "fullname": {
   "original": "nein danke ciao",
   "type": "@sys.person",
   "resolved": {
     "name": "nein danke ciao",
     "original": "nein danke ciao"
   }
 } 
}

That looks really strange for me. From my point of view it seems that there are completly wrong utterances declared as names - Does anybody know how to proceed on such a error?

[Poala_Tenorio]

It seems like there might be an issue with the way DialogFlow CX is interpreting the input "nein danke ciao" as a name. It's possible that the system entity "@sys.person" is overly permissive in this case and is incorrectly classifying non-name phrases as names.

To address this issue, you can try a few approaches:

• Custom Entity: Instead of relying solely on the system entity "@sys.person", you could create a custom entity specifically for names in German. This entity could be trained with examples of actual names to improve its accuracy.
• Review Training Data: Check the training data for your intent to see if there are any patterns or examples that might be confusing the system. Remove or modify any examples that are causing issues.
• Additional Context: Provide additional context to DialogFlow CX when training the intent. For example, if the intent is specifically for capturing names, you could provide more examples of name-related phrases to help the system learn the correct patterns.
• Feedback Mechanism: If DialogFlow CX has a feedback mechanism for incorrect interpretations, use it to provide feedback on instances where the system incorrectly identifies non-names as names. This can help improve the system over time.
• Consult Documentation or Support: Review the DialogFlow CX documentation or reach out to their support team for further assistance. They may have additional insights or recommendations for handling this issue.

By using a combination of these approaches, you should be able to improve the accuracy of name extraction in your DialogFlow CX agent.

[nimrah-waqar]

1. Review training data for patterns causing misinterpretation.
2. Verify entity annotation for "@sys.person" entity.
3. Adjust entity detection threshold in intent settings.
4. Provide additional training phrases for "@sys.person" entity.
5. Test and iterate to improve accuracy.
6. Contact Dialogflow CX support if issue persists.