Hi team, I m confused about putting training examples while getting user details. My question is, what kind of combinations that I have followed while creating examples in nlu
Under every intent, try writing as many different ways of saying the same things as possible.
ask_weather intent, it would be better to have “What’s the weather like today?” and “How cold is it here?” than “What’s the weather like today?” and “What’s today’s weather?”.
This is what I’d write as examples:
- What's the weather like today? - What's today's weather? - How cold is it here? - How hot is it right now? - Can you give me the temperature please? - Can I go outside? - Should I wear a jacket? - Is it too hot to go out? - temp plz - gimme weather - how much hot - hey tell me the temrpature
In Rasa X, you can see what conversations took place. If a user asked about the weather and the bot did not understand (for example the user said “how many celcius?”), you can add that new example to the intent by the click of a button.
Thank you so much for your reply Mr. Chrish and I have another question that why the slots are only accepting hard-coded nlu examples. For example. if I have only just two training examples with slots that are,
- may I know [muthu] (name)
- may I know [kumar] (name) In this scenario, the slot name is accepting only Muthu and Kumar. otherwise, slot doest not consider anything else. why?. can I have any chance to make the slot to be dynamic?
@Mk007 hi, as you mention the slot name as name and type hope so text so that why he is taking the only name if you create another slot by the last name then you also need to update the training example also.
Some projects for your reference, hope this will help you.Do check nlu file it will help you a lot
Yeah, only* Muthu and Kumar are accepted because names are very different from each other.
Let’s say I give you names from where I’m from, like Samir or Thérèse, if you saw these yourself you maybe wouldn’t even recognize them as names. In the same way, if I saw “muthu”, I wouldn’t know if it is a word or a name; now, because you taught me this word, I recognize it as a name.
AIs are like humans. They learn the same way as us. They just do it and respond (way!) faster
To not clutter your intents with thousands of examples, you can use Lookup Tables for your entities, like this one for names (Kumar, Muthu, Samir, and Thérèse are all already there!). Of course, the longer it is, the slower training will be.
You can read more here.
* If you type “Kamar” or “Lamar” instead of “Kumar”, it may recognize it as a name since it’s not very different. So try providing as many name variations as possible.
thank you nik
thank you chrish.