심층학습을 이용한 대화시스템의 연구방법 분석

Abstract

In this study, we discuss the basic technology of dialogue systems based on the deep neural network for natural language processing(NLP). Deep learning has become a basic technique in dialogue systems. Many researchers investigated on applying neural networks to the different components of a traditional task-oriented dialogue system, including natural language understanding, dialogue state tracking, and natural language generation, We study the recent technical advances on task-oriented dialogue systems and extent to the non-task-oriented system. We find that the end-to-end models are prevailing and representative in most of recent research papers. Also, it is blurring the boundaries between the task-oriented dialogue systems and non-task-oriented systems. In particular, the chit-chat dialogues of generative method are modeled by the end-to-end model directly. The task-oriented systems are also moving towards an end-to-end model with reinforcement learning. It is noting that current end-to-end models are still improving. We discuss some possible research directions and the technical challenges. In the future, we expect that accuracy of dialogue system will be improved further by employing reinforcement learning method.