Multi-task Joint Learning to Enhance Named Entity Recognition

Abstract Named Entity Recognition (NER) models have achieved good performance in recent years but also some shortcomings. Existing regard NER as a sequence labeling task for label prediction, without considering the impact of different stages entity recognition process on final result. can be viewed two separate subtasks: boundary detection and type prediction task. The subtasks transmit information cooperate recognition, so synergy is beneficial to NER. In this paper, we propose method split into multiple use each subtask enhance According characteristics subtasks, feature extraction extract structural useful effectively. Using extracted from or results was enhanced through gating network. We conducted extensive experiments CONLL2003 dataset, experimental show that our proposed multi-task joint learning enhances effectiveness named model.