Problem with Intent /start in telegram and train a lot of intent

Rasa Open Source
1

  1. When a Bot is started in Telegram, the system writes “/ start” and the bot responds to anything. Trying to add “/ start” as an intention in various ways “start”, “/ start” as equivalent to a greeting but the algorithm does not understand it. Always respond poorly and start a conversation giving a wrong answer

  2. I am working on the following model:

    Pre-Examen

  • Intent New Credit
    • Respond: Do you accept?
  • Intent Accept
    • Respond: Perfect. What it´s your name?
  • user respond “carlos” save in a database and next question
    • Respond: Perfect. What it´s your surname?

Should I create an intention for each answer? I simply want to store the data in a database and keep asking. How should the structure be?

  1. I am having to fail to train RASA when I have a lot of lines of code in the nlu.md file. Is it a technical or my server limitation that collapses?

Thanks

1 Like
2

Having the same thing, any telegram commands starting /something seem to bypass the entity extraction and intent classification and give an intent of “something”

1 Like
3

Any luck with this? Having the same problem

4

So “/whatever…” is taken as a command. One work around is having a pre-processor which looked for any / at the start of utterances and replaced them with your token of preference “slash whatever…” etc so it would go nicely into the RASA classifer along side our other classifer options without having to open up the RASA open source code. Then you can train on the presence of slash or your token of choice.