Knowledge and emotion integrated end-to-end dialogue method based on variational auto-encoder

Abstract

The invention discloses a knowledge and emotion integrated end-to-end dialogue method based on a variational auto-encoder, and the method comprises the steps: collecting emotion labels, dialogues, knowledge and replies, and carrying out the preprocessing of the information as training data; a model composed of a variational auto-encoder module and a copying module is built, and training is carried out; and preprocessing test data, inputting the test data into the trained model for prediction, obtaining a reply, and continuously carrying out end-to-end dialogue. An encoding module of the variational auto-encoder module encodes the emotion tag and semantic information of the input dialogue. A decoding module of the variational auto-encoder module fuses knowledge and emotions for generating content. A copy module generates a reply output in combination with the content generated by the decoder, the input dialogue and knowledge. According to the method, a variational auto-encoder structure is adopted to generate rich replies; introducing an emotion label for controlling the emotion type of the reply; information is copied from input dialogues and knowledge, so that the generated reply has richness and controllability.

Description

technical field [0001] The invention belongs to the field of natural language processing, in particular to a text generation and dialogue system, in particular to an end-to-end dialogue method based on a variational autoencoder and integrating knowledge and emotion. Background technique [0002] In 1950, Alan Turing proposed the Turing test in "Computing Machinery and Intelligence" as a way to test whether a robot can chat like a human. The Turing test can be expressed as separating the tester from the testee (a person and a machine), and asking the testee random questions through some devices (such as a keyboard). After multiple tests, if more than 30% of the testers cannot determine whether the subject is a human or a machine, then the machine has passed the test and is considered to have artificial intelligence. Turing proposed a standard for testing whether chatbots possess intelligence. It can be considered as the beginning of research on chatbots. [0003] In 1966, ...

AI Technical Summary

Problems solved by technology

[0010]The current end-to-end model can generate readable replies based on historical dialogue records, but the richness of the generated dialogue needs to be strengthened

Images