Emotional dialogue model construction method, emotional dialogue system and emotional dialogue method

Abstract

The invention discloses an emotional dialogue model construction method, an emotional dialogue system and an emotional dialogue method. According to the method, production adversarial network training is adopted to obtain a corresponding model, a generator is used for generating a dialogue reply, and a content discriminator can distinguish whether an input reply text sequence belongs to a universal reply or not and participates in the adversarial training process of the model; the emotion discriminator can distinguish whether the emotion category of the reply text generated by the generation model is a certain specified emotion category, and by adding the emotion discrimination model, the emotion category of the dialogue text generated by the generation model can be guided to be closer to the certain specified emotion category. In the man-machine conversation process, the conversation reply meeting the specific emotion can be generated, so that the reply is diversified, the emotion consistency is met, the quality of reply generated by a machine can be improved, and the user experience in man-machine interaction can be improved.

Description

technical field [0001] The invention belongs to the technical field of human-computer dialogue, and in particular relates to a method for constructing an emotional dialogue model and a related emotional dialogue system and method. Background technique [0002] As a new type of human-computer interaction, human-computer dialogue can save manpower to a large extent. It is mainly divided into task-based dialogue and chat-style dialogue in terms of functions, which are used to meet different user needs. [0003] Task-based dialogue can be used in scenarios such as smart customer service, ticket booking, and weather query in a fixed field, such as Taobao smart customer service, JD smart customer service, etc. It is necessary to understand the user's intention and dialogue based on the domain knowledge and dialogue history related to the task. management etc. [0004] Chat-style dialogues are mostly used in chat robots, such as Microsoft Xiaobing, with various conversation topics...

AI Technical Summary

Problems solved by technology

[0006] At present, the dialogue generation system mainly uses the sequence to sequence model (Sequence to Sequence, Seq2Seq), which uses the maximum likelihood estimation method to determine the generated words, which can easily lead to the selection of words with high frequency in the training data, but often such as "I don't I know" and "OK" are common sentences that lack information and have a high frequency, so it is easy to generate general answers, making the model answer relatively simple

And the model lacks emotional encoding and decoding, and does not consider the emotional relationship between questions and answers

Images