CODEWITHSUNDEEP
python-2.7svmrasa-nlu
dev
dev

Reputation: 195

RASA is not recognizing proper Intent

Used backend / pipeline: Spacy & Sklearn

Operating system: Windows 8

Issue: RASA is not recognising proper Intent

Example: I created an Intent name “GenericGreetingGM” and trained with utterances like(Good Morning, Gud mrng, gud morning, very good morning etc…). But for few utterances like (gud morning) is going into GenericSmallTalkFamily and we don’t have any words matching with “morning” or “gud” in this intent.

After communicating with few people, they suggested me to check with RASA evaluation (Intent Confusion Matrix) and below are my observations from the image.

Intent_Confusion Matrix

  • Total Utterances: 28

  • Correct Utterances: 8

  • Rest of the utterances are going to wrong Intent as I mentioned you in an example.

Below are my questions.

1. How does RASA work?

2. How do RASA analyse the text and give the Intent.

3. How can I analyse using RASA evaluation/above picture

Any help with this issue and mainly concerned with using RASA evaluation.

Thanks in Advance

Upvotes: 3

Views: 872

Answers (1)

Bhavani Ravi
Bhavani Ravi

Reputation: 2291

I have written a 4 part series demystifying the internals of RasaNLU. Hope this helps you.

https://medium.com/series/nlp-behind-chatbots-demystifying-rasanlu-318a8adb39ed

With respect to the problem, GenericGreetingGM might be because of the difference in the number of training sample under each bin. This might end up skewing the results towards a well-trained intent.

Upvotes: 1

Related Questions