High humanoid multi-mode conversation method based on neural network

Abstract

The invention discloses a high humanoid multi-mode conversation method based on a neural network, which comprises the following steps of: 1) crawling conversation information on a website, storing conversation corpus into an Elasticsearch database, and automatically generating an index; 2) training a word vector by using the crawled conversation corpus and a Word2Vec model, and converting text information of the conversation corpus into semantic expression which can be recognized by a machine; 3) searching for answer corpora through a search engine, and selecting answers sorted in the front toform an answer set; 4) respectively inputting the questions and the acquired answer set into a twin network in the conversation system; 5) introducing an attention mechanism by adopting a seq2seq deep learning model.According to the method, a traditional retrieval type conversation system and a traditional generation type conversation system are combined together, and the technical problems thatat present, most conversation robots are low in answer statement quality, high in database requirement and need to train models with mass data can be solved.

Description

technical field [0001] The invention relates to the field of chat robots, in particular to a method for constructing a highly imitative humanoid dialogue system that combines retrieval formulas and generative formulas. Background technique [0002] The types of traditional machine dialogue systems include: 1) dialogue systems based on dialogue database retrieval. This dialogue system has high requirements for the quantity and quality of dialogue materials in the database. If there are no predefined questions, you can’t make a good answer; 2) Based on the generated dialogue system, it uses the cyclic neural network to decode and encode the input sentence, and can generate new text as an answer through the existing corpus, but the disadvantage is that the answer It is too boring, and the sentences answered are prone to low-level grammatical errors such as incoherent sentences. Contents of the invention [0003] In view of this, the purpose of the present invention is to pro...

AI Technical Summary

Problems solved by technology

[0003] In view of this, the purpose of the present invention is to provide a high imitation human multi-mode dialogue method based on neural network, to solve the problem that most dialogue robots are not flexible enough to answer, do not have facial expressions, are prone to safe answers, and the quality of answer sentences is not good enough. advanced technical issues

Images