Research Of Dialogue Model Based On Deep Learning And Reinforcement Learning

The dialogue system has become one of the most common ways of human-computer interaction.Compared with other interaction,the dialogue system has the advantage of naturalness and convenient.Users without professional knowledge can also communicate with dialogue system in the simplest language.And the communication form of dialogue allows the user to free hands,so it can be applied to almost any scene.Moreover,the language is the most basic way for humans to communicate,it also makes the system with dialogue feature more friendly in use.At present,the dialogue system has been applied to fields such as smart speakers,voice assistants,intelligent customer service,virtual characters and the like.Therefore,a dialogue system with excellent semantic understanding and a large amount of knowledge has broad prospects for development.After decades of research in the academic world,dialogue technology has developed tremendously and has been widely used in the commercial field.The existing dialogue system still has a lot of deficiencies in semantic understanding,personality consistency,fusion knowledge and even the model itself.These shortcomings are also the driving force for us to explore this field.Current dialogue system will use a large number of engineering methods to solve the above problems in commercial applications,and the academic community expects to use the lower cost and smarter method to make the existing dialogue system more intelligent and easy to use.Therefore,using reinforcement learning and deep learning technology to make the dialogue system intelligent is a very challenging and urgent research field.This paper not only improves the dialogue system by using deep learning technology,but also tries to figure out how to make the dialogue more intelligence with the help of reinforce learning.This paper mainly focuses on the consistency of dialogue system,model structure,semantic understanding and knowledge fusion.The main work and innovation are as below.Firstly,the lack of personality and consistency is one of critical problems in neural dialogue systems.In this paper,we aim to generate consistent response with fixed profile and background information for building a realistic dialogue system.Based on the encoder-decoder model,we propose a retrieval mechanism to deliver natural and fluent response with proper information from a profile parts.The function of the knowledge decision part is to use the decision-making ability of reinforcement learning to select knowledge from the knowledge graph,and the function of response generation part is to generate smooth and informative response by using the background knowledge combined with context and input sentences.We apply the model to the chat conversation task that needs to be guided by the topics.The experimental results show that the proposed model can select the appropriate background knowledge to generate the response and guide the entire dialogue process according to the given topic sequence.