Improving unified named entity recognition by incorporating mention relevance

Abstract Named entity recognition (NER) is a fundamental task for natural language processing, which aims to detect mentions of real-world entities from text and classifying them into predefined types. Recently, research on overlapped discontinuous named has received increasing attention. However, we note that few studies have considered both entities. In this paper, proposed novel sequence-to-sequence model capable recognizing based machine reading comprehension. The utilizes comprehension formulation encode significant inferior information about the category. Then input sequence passes through question-answering predict mention relevance given source sentences query. Finally, incorporate BART-based generation model. We conducted experiments three type NER datasets show generality our experimental results demonstrate beats almost all current top-performing baselines achieves vast amount performance boost over SOTA models datasets.