Multi-task learning-based reply diversity multi-round dialogue generation method and system

Abstract

The invention provides a multi-task learning-based reply diversity multi-round dialogue generation method and system, a storage medium and electronic equipment, and relates to the technical field of multi-round dialogue generation. According to the invention, firstly, a multi-task learning model is constructed, the multi-task learning model comprises a pre-trained multi-round dialogue model and a VAE model, and the multi-round dialogue model comprises an utterance-level encoder, an inter-utterance encoder and a first decoder; the method also includes sequentially inputting each sentence sequence vector in the preprocessed multi-round dialogue historical information into the utterance-level encoder and the inter-utterance encoder to obtain a hidden vector containing the whole multi-round dialogue historical information; inputting the hidden vector into a first decoder, loading parameters of the first decoder from parameters of a second decoder in the VAE model, and obtaining a reply sentence sequence. VAE model decoder parameters are shared with the multi-round dialogue generation model, the decoding capability of a decoder in the multi-round dialogue generation model is enhanced, and the text quality of generated replies is improved.

Description

technical field [0001] The present invention relates to the technical field of multi-round dialog generation, in particular to a method, system, storage medium and electronic device for generating multi-round dialog with reply diversity based on multi-task learning. Background technique [0002] Applications of deep learning in multi-turn dialogue generation mainly include neural language models, sequence-to-sequence models, attention mechanisms, and hierarchical sequence-to-sequence models. [0003] Early models were mostly based on non-hierarchical structures (including neural language models and sequence-to-sequence models). However, in a non-hierarchical framework, either directly splicing historical utterances and queries or receiving utterances sequentially as input weakens the dependency between queries and replies and introduces noise. Most of the recent multi-turn dialogue generation models are implemented based on the hierarchical sequence-to-sequence model framew...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the deficiencies in the prior art, the present invention provides a method, system, storage medium and electronic device for generating multi-round dialogues with reply diversity based on multi-task learning, which solves the problem of low quality reply texts generated in existing multi-round dialogue generation. technical issues

Images