RASA high confidence on greet intent with randon input

Rasa Open Source
1

Hello folks,

I have given default greet intent data from RASA3.

nlu:
- intent: greet
  examples: |
    - hey
    - hello
    - hi
    - hello there
    - good morning
    - good evening
    - moin
    - hey there
    - let's go
    - hey dude
    - goodmorning
    - goodevening
    - good afternoon

But, I have given a random input like “change it” to the nlu. It predicts with a high confidence to the greet.

NLU model loaded. Type a message and press enter to parse it.
Next message:
change it
{
  "text": "change it",
  "intent": {
    "name": "greet",
    "confidence": 0.9899941086769104
  }

Please check and suggest me for this problem.

Note: Without using external fallback.

Thanks,

Santhosh

2

Clarifying questions @santhosh-sp

  1. Do you have other intents setup? how many
  2. Is there an out of domain intent

Looks like its going for high recall and low precision.

3

I am also facing this issue, old build is working fine but getting this issue with new release. I am using rasa 2.8.24.

4

Hi @mayukh,

  1. I have another 6 additional intents.
  2. Yes there are some intents belong to appointment and order status.
5

This is the solution for the problem.

in config.yml

   - name: "SpacyNLP"
     model: "en_core_web_trf"
   - name: "SpacyTokenizer"
   - name: RegexFeaturizer
   - name: LanguageModelFeaturizer
     model_name: "bert"
     model_weights: "rasa/LaBSE"
   - name: DIETClassifier
     epochs: 100
     constrain_similarities: true