Conversation method and system based on GRU unit hidden state constraint, and storage medium

Abstract

The invention discloses a dialogue method based on GRU unit hidden state constraint, a storage medium and a system. The dialogue method comprises the steps of converting dialogue data with emotion annotation into data in a single-round dialogue form to construct a data set; establishing a neural network model based on multiple layers of GRU units, and performing emotional constraint on the hiddenstate of the GRU units of the neural network model by utilizing emotional information in the data set to obtain the hidden state after emotional constraint; an upper-layer inheritance mechanism and alower-layer inheritance mechanism are used for the hidden state in the multi-layer GRU unit after emotion constraint, and obtaining an information fusion GRU unit hidden state; and generating probability distribution of a next word by using the GRU unit hidden state of information fusion, then optimizing parameters of the neural network model by using a back propagation algorithm to minimize a loss value, generating reply information with corresponding emotion, and realizing dialogue based on GRU unit hidden state constraints. According to the method, the reply conforming to the specific emotion category is generated according to the user input, and content consistency and emotion expression accuracy are both considered.

Description

technical field [0001] The invention belongs to the technical field of dialog systems, and in particular relates to a dialog method, storage medium and system based on GRU unit hidden state constraints. Background technique [0002] In recent years, dialogue systems have attracted much attention due to their huge potential and wide range of commercial application scenarios. On the one hand, the continuous development of natural language processing technology, such as recurrent neural network (RNN) and its variants (GRU, etc.), Seq2seq model, Attention technology, etc., provides a technical basis for the establishment of dialogue systems; The great prosperity also brought a large number of data sets, making it possible to build a data-driven, open-field dialogue system. However, it is still a very difficult challenge to build a dialogue system with correct grammar, coherent content, human-computer empathy, and reasonable responses. [0003] Studies have shown that identifyi...

AI Technical Summary

Problems solved by technology

Although there are a large number of corpus on the Internet, these corpus often contain a lot of noise, and the cost of corpus cleaning and labeling is relatively high

This restricts the development of emotional dialogue systems to a certain extent.

[0005] Second, the generated reply is difficult to take into account the consistency of the content and the rationality of the emotion

[0006] Third, simply integrating emotional information into the model does not bring satisfactory results, and requires refining and extracting emotional means, etc.

But these methods are mostly inspired by psychology, or constrained based on grammatical rules, and cannot be applied to large-scale corpora

These methods either focus on the consistency of dialogue content, or do not take into account the diversity of sentiment categories, and cannot provide satisfactory results.

Images