Research And Implementation Of Configurable Task-oriented Multi-turn Dialogue System

With the improvement of intelligence and the enhancement of securit y in recent years,the task-oriented multi-turn dialogue systems become a hotter and hotter topic.Task-oriented dialogue systems have the widespread commercial value and the prospects for development.Nowadays the task-oriented dialogue systems still meet the following challenges:1)There are no direct linkage between the intent and slots in the natural language understanding module of task-oriented dialogue system.The two tasks are still independent.2)The dialogue management module still lacks a solution that can take advantage of both deep learning models and rule matching.3)The natural language generation module of the dialogue system still needs a method to interact with the database flexibly.4)The configuration process of the existing dialogue systems is too complex,a clear configuration process is badly needed.In view of the above problems,the specific research contents of this paper are as follows:1)The natural language understanding module that establishes the bidirectional interrelated mechanism for intent detection and slot filling.In this case the two tasks can mutually promote each other.2)The dialogue manage module that combines the advantage of memory storage and neural networks.Current dialogue management methods depend on either memory or neural networks.The way that integrate the two methods can improve the performance better.3)The natural language generation module which proposes a customized method to interact with database flexibly.4)Configurable task-oriented multi-turn dialogue system.The Configuration function is added based on the dialogue system.The users can customize the action,intent and slots.A clear and customized configuration scheme is formed.Based on the above research contents,this paper implements a configurable multi-turn dialogue system that satisfies the demands of both developers and users.Concretely,it includes the following modules:1)natural language understanding module:this module has implemented a bidirectional interrelated mechanism for joint intent detection and slot filling;2)dialogue management module:this module has generally combined the advantages of memory storage and neural network.3)natural language generation:this module can interact with the database flexibly.The generated response is more reliable and humanized.4)configuration module:the user can customize the action,slot,intent and soon.The integrated configuration scheme can get rid of the cumbersome design of the existing dialogue system.