Research On Technology And Application Of Multi-turn Dialogue Based On Knowledge Graph

Dialogue system is an important branch of natural language processing,and it has been widely used in industry in recent years.Dialogue system can not only help people to complete specific tasks,but also answer specific knowledge and chat with people.In the communication between people and machines,a lot of knowledge is often involved.In order to make the machine response more rich and meaningful,it is very important to integrate the knowledge graph into the multi-turn dialogue.Multi-turn dialogue based on knowledge graph is a multi-turn dialogue with knowledge graph as the center,using entity discovery and linking technology to integrate multiple information together.This paper focuses on two tasks: multi-turn dialogue system based on knowledge graph and entity discovery and linking,and this paper develops intellectual dialogue in the field of Winter Olympic Games based on these methods.Firstly,in the multi-turn dialogue based on knowledge graph,this paper mainly uses the end-to-end method to research this task.In the process of analyzing the baseline system,we find that the baseline system don't use the dialogue history information effectively and don't model the knowledge graph reasonably.To solve the problem of using dialogue history information,this paper uses hierarchical modeling method to model dialogue history,and on this basis,this paper uses word level and sentence level attention mechanism to use dialogue history information effectively.To solve the problem of knowledge graph modeling,this paper tries to use Trans D to vectorize the knowledge graph,and learn the entity and relationship information in the knowledge graph,thus improve the effect of knowledge utilization in the dialogue.The experimental results show that the hierarchical modeling method and Trans D can improve the effect of generating response in dialogue system effectively.Secondly,in the task of entity discovery,in order to solve the problem of data sparsity,this paper proposes a named entity recognition model of mixing char and word information,and use convolutional neural network to further extract context information to improve the effect of entity discovery.In the task of entity linking,this paper integrates the historical information of dialogue and use attention mechanism to improve the effect of entity disambiguation due to the continuity and information omission in multi-turn dialogue.The experimental results show that the proposed methods are more effective than the baseline models.Finally,this paper applies the research results to the intellectual dialogue in the field of the Winter Olympic Games,designs and implements a multi-turn dialogue system based on the knowledge graph of the Winter Olympic Games.In the process of system implementation,this paper proposes a semi-automatic data generation method based on knowledge graph,which effectively solves the problem of lack of dialogue data in the field of Winter Olympic Games.It shows that the relevant research has a high application value.