Intent Discovery uses the ITSM Virtual Agent NLU model?

[Lynda1]

We created our own NLU model (I do not understand why) and we just upgraded to San Diego, running analysis on the Intents,  It appears the results are using the ITSM NLU for  Virtual Agent module, it appears to ignore the module we created.

Topic recommendations appears to use Topics in this scope only  ITSM Virtual Agent Conversations as well.

Am I seeing this correctly?  If so, I have some work to do!  Just wanted to confirm I am reading this right.

[Brian Bakker · Accepted solution]

Hello Lynda,

You should create your own custom NLU Model, as the out-of-box custom NLU Models are read-only and designed to be used as templates, so you cannot customize them.

The Intent Discovery report is based on Taxonomy and ITSM is the only taxonomy supported in San Diego, but more will be added in Tokyo, hence you will only see Topic recommendations based on this scope only. You can change it to "No taxonomy" for generic Topic recommendations -

In the generated Intent Discovery report, you can add the intents it discovered to a new NLU Model, by using the Add to model button. It does not list existing NLU Models and so you can create a new NLU Model for all the intents you want to use in your own custom NLU Model, as you can import them, as shown later -

You provide the name of the new NLU Model, such as "ITSM Model" -

Then you add all the required Intents to the new NLU Model -

Now you can import these intents into your own custom NLU Model -

Once imported, you can add more utterances where required, as all the intents in an NLU Model should have an evenly distributed number of sample utterances. Train the NLU Model and then test it using the NLU Batch Testing tool -

I hope this helps.

Regards,

Brian

[Lynda1]

I realized the need for custom models right after I typed this out. Temporary brain lapse with the excitement of this new feature!

I am going to try the "no taxonomy" see what it does.  It is just annoying that the findings are on ITSM Intents when I have topics and intents that catch the utterances already. I notice these analysis are used in the Topic recommendations as well.

Thank you for the detailed response! It gives me a way to get what I want out of these analysis's!