Design And Implementation Of Task-Oriented Dialogue System Platform

With the advent of artificial intelligence,intelligent chat robots have become a hot topic.People want machines to think like humans,talk to humans,and become human helpers.In the chat robot,personal assistant robot is one of the most important type.This type of chat robot is often called task-oriented dialogue system.Taskoriented dialogue system is mainly to help people to complete the instructions or commands such as checking the weather,booking a ticket,based on natural language interaction with users.A task-oriented dialogue system mainly consists of two modules: language understanding module and dialogue management module.The main task of language understanding is to convert unstructured natural language commands from users into structured semantic representation,used as operation parameters of the target action.The main task of dialogue management module is to maintain session context of each user,interact with users based on natural language and complete users' target intent.First of all,from the aspect of the vertical dimension of task-oriented dialogue system,this paper focuses on the accuracy of language understanding.We investigate the method of language understanding based on the method called slot filling.The slot filling method of language understanding can be regarded as the sequence labeling task of natural language,for that the task input is the instruction type statement from a user,the output of slot filling task is semantic entity types on each words,such as the date of departure,the place of departure.So slot filling task can be regarded as similar to the task of named entity recognition task,and we can treat it as sequence labeling task.In this paper,we extract the semantic slots based on deep learning methods,and the experiment results show that the method based on deep learning is greatly improved compared with the baseline method based on traditional statistical machine learning methods like CRF.Secondly,from the aspect of the horizontal dimension of the robot system,we design and implement a platform for creating task-oriented dialogue system.This platform offers developers the language understanding algorithms and the dialogue management as a black box,enabling developers to easily develop their own taskoriented dialogue system to execution the specific intent commands,as long as the developers upload corpus and define the parameters of semantic slots of the system.