A Tree-Based Approach for Building Efficient Task-Oriented Dialogue Systems

• Published in: Artificial Intelligence for Communications and Networks Vol. 396; pp. 526 - 536
• Main Authors: Gan, Tao, Li, Chunang, Xi, Yuhui, He, Yanmin
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2021; Springer International Publishing
• Series: Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering
• Subjects: Data-driven; Intent-slot tree; Rule-based; Task-oriented
• ISBN: 3030901955; 9783030901950
• ISSN: 1867-8211; 1867-822X
• DOI: 10.1007/978-3-030-90196-7_45

Task-oriented dialogue systems have attracted increasing attention. The traditional rule-based approaches suffer from limited generalization ability as well as the high cost of system deployment, whereas the data-driven deep learning approaches are data-hungry and the domain-specific data is insufficient for full training their models. In this paper, we present a hybrid method which combines the strengths of both rule-based and data-driven approaches. We first establish intent-slot trees from the standard multi-turn dialogue corpus in specific domain. During the dialogue, the power of deep language understanding model is exploited to enhance the generalization ability of the system and the multi-turn dialogue proceeds following the path of the intent-slot tree established. Experimental results show that the proposed approach achieves superior performance over deep learning ones which demonstrates its effectiveness in building task-oriented dialogue systems under a limited amount of training data.