Dialogue generation method and system based on bidirectional asynchronous sequence

Abstract

The invention discloses a dialogue generation method based on a bidirectional asynchronous sequence, and the method comprises the steps: S1, obtaining a dialogue text, recognizing a keyword in the dialogue text, and generating a word vector corresponding to the keyword; s2, generating a statement vector of the dialogue text based on the word vector obtained in the S1; s3, calculating the similarity between the statement vector of the dialogue text obtained in the S2 and a statement vector of a sample question in a knowledge question and answer library, and generating an answer set comprising a plurality of candidate answers and word vectors corresponding to the candidate answers; s4, based on the answer set generated in S3 and the corresponding word vector, obtaining an initial answer text through a bidirectional asynchronous sequence algorithm; and S5, correcting the initial answer text, and outputting a final answer text. The invention further provides a system for implementing the method. The answer text generated by the method carries more extended contents, the problem of repeated word use is avoided through the correction algorithm, and the quality of the answer text is further improved.

Description

technical field [0001] The present invention relates to the fields of artificial intelligence, neural network and natural language processing, in particular to a dialogue generation method and system based on bidirectional asynchronous sequences. Background technique [0002] Human-computer dialogue has always been a hot research field in natural language processing. Diversity and uncertainty are the key factors for a more "real" human-computer dialogue. Thanks to the development of neural networks, data-driven generative dialogue models are more flexible It shows great potential in terms of degrees of freedom and interaction. [0003] Before neural networks are widely used, the main methods of dialogue generation tasks rely on statistical methods and retrieval methods, which are not only limited to specific fields, but also require artificially set rules for guidance in most tasks. After the emergence of neural networks, especially after the widespread application of convo...

AI Technical Summary

Problems solved by technology

[0004] However, the performance of the existing data-driven neural network dialogue generation model is not good, mainly due to three problems:

[0006] (2) The encoder cannot capture enough knowledge information to provide to the decoder for decoding

[0007] (3) Due to the complexity of human languages, the quality of probabilistic generation is usually difficult to meet the requirements of language semantics

This method can only identify the corresponding reply sentence for the question sentence, and the output reply sentence may cause repeated words in the output sentence because the sampling results are too similar.

Images