Self-selected generation retrieval framework based on topic keyword model

Abstract

This paper presents a self-iterating framework implemented in a retrieval dialogue scenario that captures the one-to-many mapping relationships in multiple rounds of dialogue and retrieves multiple conversational contexts from the dialogue corpus that are both semantically diverse and logically coherent. The training model of the framework also allows for the selection and optimisation of individual models in the framework to automatically expand the dialogue corpus to form a one-to-many dialogue dataset.