Transfer learning for a Vietnamese dialogue system

Training data plays an essential role in the neural network-based dialogue system. Although most of the recent works researching about dialogue system base on a deep neural network show a significant improvement to traditional methods, they required many labeled data for both training and evaluating. In this paper, we present the result of using transfer learning in a Vietnamese dialogue system for the classification task, and we further show benefits of transfer learning on named entity recognition task by experiment results on public VLSP dataset and meets the performance of training from scratch with only 10% of training data.