Is there a way to make some words in a training example count more than others when Rasa looks for intents? If I put them in twice in the example, will they have more weight on determining what the intent is?
Here’s a typical training example:
how did your thinking change :: **ask, tag500,**
ask and tag500 are keys I want to associate with this example, so I could do this:
how did your thinking change :: **ask, tag500,** **ask, tag500,** **ask, tag500,**, **ask, tag500,**
but would it have any effect on favoring examples with those tags be identified as a given intent?
Alternatively, I could flag them as entities, even though I don’t need them as entities, if that would raise their importance:
how did your thinking change :: [tag1](ask), [tag2](tag500)
@bferster can you please share some examples of some words in the training example? If you emphasize one word in the training example it will not impact the intent I guess.
Repeating the keyword multiple times in the same example may lead the bot to believe that it usually appears that many times in a user input, which may thus lead to the opposite of the desired effect.
It would be better to have multiple examples containing the keyword.
For example, for a want_pizza intent:
Bad:
- I want pizza pizza pizza pizza
Good:
- I want pizza
- Can I get pizza?
- I need to eat a pizza
- Pizza would be amazing to have right now
Examples should be as close to what a user may type as possible.
That’s good to know.
I’m already tagging multiple events, but I want the tags to mean more than just another word. I’m training on 4,000 remarks made by live teachers, hand-coded and tagged to 5 intents
Would making it an entity help, like this?
how did your thinking change :: [tag1](ask), [tag2](tag500)
