Dialogue Act Classification Model Based on Deep Neural Networks for a Natural Language Interface to Databases in Korean
Minkyoung Kim, Harksoo Kim · 2018
Dialogue act classification is an essential task for implementing a natural language interface to databases because speakers' intentions can be represented by dialogue acts (domain-independent speech act and domain-dependent predicator pairs). To resolve ambiguities in dialogue act classification, various machine learning models have been proposed over past 20 years. In this paper, we propose a dialogue act classification model using a convolutional neural network and a long short-term memory network. The proposed model generalizes an input utterance into an embedding vector by using the convolutional neural network. Then, it annotates utterance sequences with dialogue act labels by using contextual information based on the long short-term memory network. In the experiments with a Korean goal-oriented dialogue corpus, the proposed model showed better performances than the previous models.