Models Distillation with Lifelong Deep Learning for Vietnamese Biomedical Named Entity Recognition

Abstract

In realistic data, named entities may appear in a variety of rich contexts with unique characteristics and the performance of the named entity recognition (NER) task directly affects other NLP problems. Although both lifelong learning and deep learning have proven effective in many problems, including NER, the suitable combination of these two research directions is still limited. This paper describes a lifelong deep learning model for Vietnamese Biomedical NER based on model distillation mechanism. Our approach achieves potential results, helps to boost 2.16% compared to original deep learning model.