Research And Implementation Of Emotional Text Generation Technology In Multi Round Dialogue

In recent years,with the development of big data and neural networks,the research of open-domain dialogue systems has gradually attracted people's attention.Researchers hope that the dialogue system can conduct smooth and emotional dialogue with humans.Studies have found that incorporating emotional factors into the answers of the dialogue system can improve users'satisfaction with the system,as well as the quality of the dialogue system's answers and the ability to solve problems.The study of emotional dialogue,which can make machine-generated responses with emotions,is the current research hotspot of dialogue systems.At present,most of the researches on emotional text generation in dialogue system focus on how to integrate the specified emotion into the sentences generated by the system.when the model talks with the user,the user needs to specify the emotion category of the response generated by the model,which affects the fluency of the dialogue between model and user.To solve this problem,our paper focuses on the emotional changes in multi round dialogues,and proposes a method of integrating emotion into multi round dialogues without specifying the emotion of generated response.This model is based on the ECM model.Firstly,the emotion prediction model encodes the user's input into a sentence vector through the encoder,and the output is the emotion category of the response generated by the model.The emotion category will be used as the input of the ECM model to generate the response of the predicted emotion category,which solves the problem that the ECM model needs the user to specify the model to generate the emotion category of the response.We use LSTM to model the emotion change in multi round dialogue.The effect under different parameters is compared through experiments.It is determined that we can get the best resutl when the word vector dimension is 64,the hidden vector dimension is 256,and the batch size is 64.The accuracy of the model is 82.4%.On this basis,we also extract the context information in the dialogue,and the context information vector will also be input into the ECM model;when the decoder generates the response,it adds the context information vector and the emotion state vector into the decoder,and the decoder will eventually generate the response corresponding to the emotion category,which has strong relevance to the context.Experiments show that compared with the most basic end-to-end model,the confusion degree of the proposed method is reduced by 1,and the accuracy is improved by about 40%.Moreover,compared with most systems that need to manually select the model to answer emotions,the system in this paper is more in line with the habit of daily conversation.At the same time,this paper designs and implements a multi round dialogue system with emotion.According to the user's input,the system can generate emotional answers.The system is divided into the following modules:emotion prediction module,context information extraction module,emotion state processing module,database module,dialogue system main body module.After that,the flow and function of each module are introduced in detail,and the system is tested.The test results show that the system can complete the training and dialogue functions normally,the user can communicate with the system normally,and the system can generate the corresponding emotional response according to the context.And compared with the traditional end-to-end model,the response emotion generated by this model is more obvious and context sensitive.