What are the best resources to fetch data for training NLU?

[surendra_koritala]

@karen-white: Dear Karen,

This might be silly question. But, I am having trouble with training the model. what are the best resources to fetch data for training NLU.

I am having hard time with creating intents for each and every possible scenario.

Regards, Surendra

[karen-white]

Hey @surendra_koritala, just wanted to clarify - did you want to ask this question to Dr. Catherine Breslin as part of our AMA, or was this a troubleshooting question about Rasa Open Source?

As a rule of thumb, we’d recommend not necessarily creating intents for every possible scenario - instead, start with the intents you need to support your bot’s basic features and add new intents when users actually start to express those intents in conversations with your assistant. Another best practice is to make sure your intents aren’t too similar. For example, we recommend using a single intent:inform to capture all of the scenarios where a user is providing information and only the entity is important (instead of intent:provides_name, intent:provides_address etc). And although this blog post is a bit out of date in terms of our recommended pipeline components (we’d now recommend DIET), you might find the best practices in the second half of the post helpful : Rasa NLU in Depth: Intent Classification

[Emma (Emma Wightman)]

Hi @surendra_koritala

Thank you for clarifying your question, I have moved it to the AMA thread so that Catherine can answer it live tomorrow June 4.