A Pipeline Dialogue System Scheme

Conference: ICMLCA 2021 - 2nd International Conference on Machine Learning and Computer Application
12/17/2021 - 12/19/2021 at Shenyang, China

Proceedings: ICMLCA 2021

Pages: 5Language: englishTyp: PDF

Personal VDE Members are entitled to a 10% discount on this title

Authors:
Sun, HaoYuan; Zhao, ChunYi; Liu, ShiQiang; Jiang, HuanHuan (Shenyang Institute of Computing Technology, Chinese Academy of Sciences, Shenyang, China & University of Chinese Academy of Sciences, Beijing, China)

Abstract:
Task-oriented dialogue is a current research hotspot. It usually consists of three modules, namely the dialogue state tracking module, the strategy module, and the natural language generation module. In terms of implementation, there are pipeline and end2end methods, although the end2end model structure is simpler, the effect is also good on some datasets, but there is no outstanding industrial-grade product. We believe that an excellent artificial intelligence product must be composed of a neural network model and symbolic rules. The model built by Pipeline can just use symbolic rules to process the generated results of the model, making the dialogue system controllable and more stable. In this article, based on the SGD dataset, we first proposed a dialogue system solution built in Pipeline, which has a good experience in manual testing.