Research And Application Of Multi-Task Hybrid Dialogue System Technology

In task-oriented dialogue systems,users’ needs often come from many different domains,and the domains involved may be related to each other,so how to design a task-oriented dialogue system that can handle multiple domains at the same time,i.e.,multi-task hybrid dialogue system,becomes a problem worth studying.There are still three difficulties in multi-task hybrid dialogue:1)the scale of dialogue ontology is too large and relies strongly on complete ontology knowledge;2)it is difficult to track implicitly mentioned slot values because it is difficult to model the relation information between domains and slots in multi-task hybrid scenario explicitly and efficiently;3)the information in the dialogue history that is truly task-related is sparse,which is not conducive to guiding the system to generate responses that can accomplish the task.To solve the above problems,this paper proposes a multi-task hybrid dialogue model,and the main work and contributions can be summarized into three points:1)To cope with the problem of large ontology knowledge,the model uses a copy mechanism to generate dialogue states and system replies in the decoding phase,so it does not rely on full ontology;2)The model introduces a hierarchical task graph that can explicitly and efficiently model the relation between domains and slots,and it uses a graph attention mechanism for modeling,which improves the model’s ability to track implicitly mentioned values of the slots;3)the input of the response generation phase uses generated dialogue states,which has high information density,instead of the complete dialogue history in order to make the system response generated by the model more focused on completing the task.Experimental results on large-scale multi-task hybrid dialogue dataset show that the model designed in this paper outperforms the best baseline model in both the tasks of dialogue state tracking and dialogue response generation.Further case studies and comparative experiments show that the model is able to capture the implicitly mentioned slot values when generating dialogue states and generate fluent system response which can help to complete the task.Finally,based on the proposed model,a multitask hybrid dialogue system for multi-domain information query and reservation is designed and implemented in this paper,which can provide efficient and convenient dialogue services for users.