Can Transformer Models be used for Training Chatbots?
Note - I am talking about the transformer model google released on the paper 'Attention is all you need'
Data Science Stack Exchange is a question and answer site for Data science professionals, Machine Learning specialists, and those interested in learning more about the field. It only takes a minute to sign up.Sign up to join this community
A transformer is just a neural network. Sure, it is way more complex than a feed forward one, but it still is a neural network. As long as you provide the correct dataset (supervised case, so the correct pair input-target) your model should be able to learn the hidden representation that make it able to answer to new questions.
I found this interesting tutorial, try to check it out.
The Transformer model is a sequence-to-sequence model, that is, it is meant to address problems where the input is a sequence of discrete tokens (i.e. text) and the output is also a sequence of discrete tokens.
Therefore, a Transformer is well suited to be trained with a dataset of dialogs where the input is a statement or question and the output is the answer. This is usually called a "chit-chat" chatbot, because they are not backed by a knowledge base. They can just have "small talk".