A Multi-task Framework for Named Entity Recognition

Existing models usually treat named entity recognition (NER) as a single-target learning task. However, the NER task contains two prediction targets: boundary and type, the single-target learning can limit the performance and efficiency of the model. To solve this problem, we examine the NER task from a new perspective, and proposes a multi-task learning framework. The proposed framework converts the NER task into two subtasks: boundary labeling and type labeling, and then joint learning of the subtasks. Finally, the learning results of the subtasks are combined to establish the learning model of complex tasks. Extensive experiments were conducted on two publicly available datasets, the results show that our proposed multi-task learning model can be flexibly combined with various excellent semantic coding models to transform the single-target task into a multi-task learning mode, and our work can greatly decrease the learning time and bring huge performance improvement.