Research On Dialogue Generation Method Based On Word-Level Weights And Adversarial ECM Model

In recent years,with the rapid development of computer performance,especially the use of graphics processing units(GPU),the research on the field of natural language processing has been promoted.Among them,the dialogue system based on deep learning has made great progress.However,the current research on dialogue systems focuses on the understanding of sentence semantics and does not take into account the impact of emotional factors on the dialogue system.Emotional quotient(EQ)is an important part of human intelligence.People with high EQ can understand the emotion of the statement and give the best response according to the emotional factors in the dialogue.So,allowing the dialogue system to express emotions can improve the quality of conversations generated by the dialogue system,resulting in a better user experience.At the same time,the use of the generative adversarial networks(GAN)in the dialogue system improves the accuracy of the traditional machine learning method and makes the generated dialogue more humanized.Firstly,this paper studies the structure and principle of the machine translation model Transformer and the language sub-network in GNA-RNN model.In the Transformer model,the multi-scale transformation of the word vector can obtain the semantic features of different dimensions,which improves the accuracy of the model.The GNA-RNN model increases the weight of important words in the sentence through the language sub-network,thus improving the ability to search for related images.In order to improve the accuracy of the sequence generation model,this paper proposes a word-level weights network,taking the sentence vector embedded in the word as input,the weight of each word in the sentence as the output,and multiplying the generated weight by the statement generated by the sequence generation model to obtain the final Output.By increasing the weight of important words in the sentence,the influence of the sequence in the generation of the sequence is improved,thereby improving the accuracy of the sequence generation model.Secondly,this paper analyzes and studies the multi-emotional dialogue generation model ECM and the sequence generation model SeqGAN.Although the original ECM can generate statements with emotional factors,the generated partial statements have syntax errors and inaccurate semantic expressions.SeqGAN solves the shortcomings of syntax errors and statement redundancy in traditional dialog systems.In this paper,the SeqGAN model is combined with the ECM model with word-level weights,and a new multi-emotional dialogue generation model ECGAN is proposed.The model uses ECM as a generator and hierarchical RNN as a discriminator.ECGAN can generate more humanized statements and improve the accuracy of the original ECM model.Finally,the ECGAN model is trained using the NLPCC2017 dataset and compared with the previous ECM model and the ECM model with word-level weights.The comparison results show that the sentence generated by ECGAN is more accurate and more humanized.