SNER-CS: Self-training Named Entity Recognition in Computer Science

Abstract As the number of scientific publications grows, especially in computer science domain (CS), it is important to extract entities from a large CS publications. Distantly supervised methods, generating distantly annotated training data by string match with external dictionary automatically, have been widely used named entity recognition task. However, there are two challenges use methods NER One that more and new tasks, datasets proposed rapidly, which makes difficult build knowledge base high coverage. The other noisy annotation, because no uniform representation standard domain. To alleviate problems above, we propose novel self-training method based pretraining language model label automatic construction system (SNER-CS). Experimental results show SNER-CS performs previous state-of-the-art