Today all the training comes from an YAML file.
If the users try some interaction against the bot that it's not trained to answer, the bot will return an error message, and nothing else.
The supervised learning skill is based on the ability of learn from it's mistakes.
So here what I'm thinking is to name a supervisor user to bot, and in every error, it sends a message to the user asking what it should respond, and then take the expected message and the answer to a new interaction node, and retrain the classifier with that new interaction node.
Today all the training comes from an YAML file. If the users try some interaction against the bot that it's not trained to answer, the bot will return an error message, and nothing else.
The supervised learning skill is based on the ability of learn from it's mistakes. So here what I'm thinking is to name a supervisor user to bot, and in every error, it sends a message to the user asking what it should respond, and then take the expected message and the answer to a new interaction node, and retrain the classifier with that new interaction node.
Any ideas?