LADA-Trans-NER: Adaptive Efficient Transformer for Chinese Named Entity Recognition Using Lexicon-Attention and Data-Augmentation

Recently, word enhancement has become very popular for Chinese Named Entity Recognition (NER), reducing segmentation errors and increasing the semantic boundary information of words. However, these methods tend to ignore relationship before after sentence integrating lexical information. Therefore, regularity length not been fully explored in various word-character fusion methods. In this work, we propose a Lexicon-Attention Data-Augmentation (LADA) method NER. We discuss challenges using existing incorporating NER show how our proposed could be leveraged overcome those challenges. LADA is based on Transformer Encoder that utilizes lexicon construct directed graph fuses through updating optimal edge graph. Specially, introduce advanced data augmentation obtain representation task. Experimental results done can considerably boost performance system achieve significantly better than previous state-of-the-art variant models literature four publicly available datasets, namely Resume, MSRA, Weibo, OntoNotes v4. also observe generalization application real-world setting from multi-source complex entities.