Research And Implementation Of Dialogue State Tracking Based On Dialogue History Structure

Dialogue state tracking is the key of pipeline task-oriented dialogue system,which provides decision basis for dialogue policy.Dialogue state refers to the set of task-related slot value pairs obtained by the current dialogue,and the main task of dialogue state tracking is to continuously obtain the dialogue state accumulated at the current dialogue moment.Dialogue state tracing needs to model the mapping from the dialogue history to the dialogue state,so modeling the dialogue history is its core.Based on the review of the existing work related to the t dialogue state tracking,this paper studies and applies the insufficient modeling of the structural information of dialogue history.The main contents include the following two aspects:First,this paper presents a dialogue state tracking model based on dialogue history structure,not only uses the structural relationship between the system and the user in the dialogue to construct the usefulness of the dialogue history,but also models the dialogue history hierarchically and describes the semantic vector of the dialogue history composed by each part of the dialogue history.The preceding operations make fine-grained use of historical dialog information and reduce the impact of redundant information on slot value generation.In addition,by modeling the relationship between slots and dialogue history,a more effective representation method of dialogue history is obtained.Experimental results on MultiWOZ2.0 show that the proposed model can effectively alleviate the influence of redundant information in dialogue history,and in the same type of small-scale model without the use of pre-training model,the current best performance is achieved,and the parameter scale is only about 10%of the large model to achieve more than 90%of the performance of the large model,which has a comprehensive advantage in practical deployment.Secondly,a human-machine dialogue demonstration system is designed and implemented as an intelligent travel assistant.The system is implemented according to the software engineering method,and the dialogue state tracking model based on the dialogue structure proposed in this paper is combined with the modules of dialogue policy and natural language generation.Finally,the test proves that the system can meet the basic requirements and provide effective human-machine dialogue service in the restaurant query and other tasks.