Research And Design Of Infant And Toddler Service Platform Based On Task-based Multi-turn Dialogue

With the development of society,parents in the new era have high requirements for infant and young children’s services.How to make infants and young children grow up healthily has become a social topic.In this paper,the task-based multi-turn dialogue system based on pipeline is deeply studied,and it is modeled and applied to the field of infant and young children’s services,and finally the infant and young children’s service platform is successfully developed.This platform was developed to solve some of the problems parents have encountered in raising infants and young children.The task-based multi-turn dialogue system aims to complete the specified task by obtaining effective information in multiple rounds of interaction with users according to specific service requirements.The dialogue system can be divided into single-round dialogue and multi-turn dialogue according to the dialogue rounds.Compared with single-round dialogue,multi-turn dialogue has more problems on the context of human-machine dialogue,the completion of reference omission,and the clarification of complex requirements.The complex requirements are also closer to the application scenarios of chatting,consultation,recommendation,and service needs in real life,which are the research focus of this paper.The main work of this paper is as follows:(1)This paper proposes an intent recognition model based on external memory.The intent recognition of the dialog system determines the overall direction of the dialog,and the subsequent dialog will only be meaningful if the system correctly recognizes the user’s true intent.The intent recognition model proposed in this paper is based on the Bidirectional Long Short-Term Memory Network(Bi LSTM),and introduces an external memory unit and a self-attention mechanism.The external memory unit has a large storage capacity,which can solve the problem that the internal memory unit is easy to lose key information due to its small storage capacity;the self-attention mechanism can capture more semantic information,which is conducive to the recognition of intentions.(2)This paper proposes a dialogue policy model based on DQN.The traditional maximum likelihood estimation-based model generates sentences with the greatest probability for each reply.These sentences are often universal replies.When the questions and replies are similar or the same,the dialogue system is prone to fall into an infinite loop;the dialogue in the dialogue system The strategy module can introduce the reinforcement learning algorithm Q-Learning.However,the dialogue system usually has a large state space or action space.Q-Learning needs to establish a Q table to maintain,which will consume a lot of memory.The dialogue strategy module introduces the deep reinforcement learning algorithm DQN to solve the above problems.The DQN algorithm guides the selection of responses in each round of the multi-turn dialogue process.The principle of each selection is to maximize future rewards,introduce new information to the dialogue,and improve the dialogue.system performance.(3)In order to solve some problems encountered by parents in the process of raising infants and young children,this paper develops a service platform for infants and young children based on task-based multi-turn dialogues.The core technology used in the development of this platform is the model studied in this paper and the pipeline-based task-based multi-turn dialogue related algorithm.Based on social issues,this paper conducts in-depth research on the development background and goals of the platform;then analyzes the needs of the platform and designs the system,and determines the overall framework of the platform;finally,the infant service platform is successfully developed.