Recent Advances and Challenges in Design of Non-goal-Oriented Dialogue Systems

摘要

Human being is pervasively surrounded by smart devices that provide numerous services to them. These devices are equipped with interfaces that are natural and intuitive with the aim of providing user an effortless and organic interaction with the devices. A dialogue agent is one such interface that interacts with the user in natural language. Recent paradigm classifies them as goal-oriented and non-goal-oriented dialogue systems. The aim of goal-oriented dialogue systems is to assist the user in completing a task. Evidently the design of goal-oriented-dialogue agents has made a lot of progress. But the interactions with non-goal-oriented dialogue systems are reasonable, open- domain and more applicable to real-world applications. This paper reviews the state-of-the-art in Non-goal-oriented dialogue systems. The design of such systems has advanced due to Big Data hence most of the recent models are data-driven. This paper is a comprehensive study of data driven systems - advances in learning models and recent frontiers. Also provide an insight on datasets and evaluation methods and the limitations that the data-driven methods, datasets and evaluation methods present.