If a user's turn is a query *q*, we must retrieve and repeat another turn *r* as the response from a corpus *C*. The *r* must be chosen based on its similarity metric to *q*, which is calculated with the below equation.

Response by Retrieval

Response by generation is an encoder-decoder task that learns from a corpus to transduce a user's prior question into an acceptable answer. Basic encoder-decoder models tend to yield very simple responses like "Okay" or "I'm fine," but one way to work around this is to fine-tune a large language model on a conversational dataset and using the model to directly generate responses.

Response by Generation

Response generation systems are a corpus-based chatbot's methods of creating responses via retrieval or generation methods.

University of California, Santa Cruz

Corpus-based chatbots are chatbots that draw their conversational data from corpuses, or vast linguistic datasets, of human conversation from various sources like social media and TV shows. 

Corpus-Based Chatbots

An on-going but a helpful book resource about NLP
https://web.stanford.edu/~jurafsky/slp3/

Learn Before

Related

Learn After