Multi-round dialogue generation method and system based on information enhancement

Abstract

The invention provides a multi-round dialogue generation method and system based on information enhancement, and relates to the technical field of multi-round dialogue generation. The reply is generated through a pre-constructed multi-round dialogue model, the multi-round dialogue model comprises a multi-dimensional word level encoder, a speech level encoder and a decoder, and the reply generation process comprises the steps that multi-round dialogue historical information is acquired and preprocessed; the preprocessed multi-round dialogue historical information passes through a multi-dimensional word level encoder based on a multi-information embedding mechanism, and sentence vectors are obtained; inputting the sentence vector into an utterance level encoder to obtain an utterance level encoding vector; and decoding the utterance-level coding vector through a decoder to generate a reply. According to the method, context information is comprehensively captured through a multi-information embedding mechanism, and more relevant context information is generated to generate replies.

Description

technical field [0001] The present invention relates to the technical field of multi-round dialogue generation, in particular to an information enhancement-based multi-round dialogue generation method, system, storage medium and electronic equipment. Background technique [0002] Most of the existing multi-round dialogue generation methods are implemented based on the hierarchical sequence-to-sequence model framework, which can be summarized as RNN-based hierarchical models and Transformer-based hierarchical models. Hierarchical models based on RNN usually filter contextual information through attention mechanism or similarity, but RNN has the problem of position deviation, usually pays more attention to the short-distance context, and cannot better learn the long-term dependencies between contextual information in the dialogue, and the generated Dialogue lacks coherence. Based on Transformer's hierarchical model, due to the introduction of multi-head attention mechanism, i...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the deficiencies of the prior art, the present invention provides a multi-round dialog generation method, system, storage medium and electronic device based on information enhancement, which solves the technical problem that the existing multi-round dialog generation method cannot fully capture context information

Images