Hello, I have configured lex bot with an intent that consists of one sample utterance. While testing, if user utterance is lengthy, it triggers fallback intent even though it is matching with configured intent with confidence score of '0.78' which is greater than default threshold(0.4). Help me with this issue.
Steps to Reproduce:
- Create a custom bot in lex v2 console with default configuration and a intent named "PolicyChange".
- Add this sample utterance - "I want my husband's name removed from the policy".
- Configure fulfillment successful message for that intent with a message something like - "Ok we will update policy"
- Build the bot and test it in test console with utterance - "My husband died last month due to covid. Now I want his name removed from the policy. Can you help me with that."
- Inspect the JSON input and output response in test console. You can see "FallbackIntent" is triggered. But, in Interpretations the "PolicyChange" intent has confidence score of "0.78".
AWS 官方已更新 2 年前
Have you thought about invoking a Lambda function w/ a python script to interpret the utterance? I know that's more complicated but it's worth a try.
Hi Jennifer, that is the only option I am left with but ideally Lex itself provides interpretations before lambda invocation and it supposed to interpret the intent with highest confidence score, but it gives unexpected results. Is it a normal lex reponse behaviour or something wrong here in this case?