Optimization of Intention Detection Based on Metric Learning

Abstract

With the development of machine learning, transfer learning has great development prospect and commercial value compared with the traditional supervised learning. As neural network developed, transfer learning based on metric learning is widely used in the field of Computer Vision and gradually applied to Natural Language Processing. This paper proposes to use BERT encoder and BiLSTM to improve the performance of intention detection especially in classification performance. SMP2017 data set shows that it can effectively improve the accuracy of intention detection when the sample size is small and uneven.