2

I was going through the Watson knowledge studio and NLU. I understood that WKS is supervised learning because we have to trained WKS with pattern and taxonomy to get the desired result. Can some body help me to understand about Watson NLU, Do we not trained the NLU and it returns the result based on internal implementation. Does NLU trained itself ?
Is Watson NLU supervised and unsupervised learning API ?

Thanks

Alex Bravo
  • 1,601
  • 2
  • 24
  • 40
Ritesh Kumar
  • 23
  • 3

2 Answers2

3

NLU returns results using models created by IBM, but you can use a custom model from WKS to further customize the service in order to identify domain-specific entities and relations.

https://console.bluemix.net/docs/services/natural-language-understanding/customizing.html#customizing

Felipe Paixao
  • 411
  • 3
  • 7
1

Felipe's answer above is correct. These are both instances of SUPERVISED learning. NLU does NOT do any UNSUPERVISED learning - where it learns from whatever the users input. This is by design - unsupervised learning can lead to some pretty scary results (see MS Tay Chatbot - https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist)

Daniel Toczala
  • 522
  • 2
  • 15