BERT-Based Transfer-Learning Approach for Nested Named-Entity Recognition Using Joint Labeling

Named-entity recognition (NER) is one of the primary components in various natural language processing tasks such as relation extraction, information retrieval, question answering, etc. The majority research work deals with flat entities. However, it was observed that entities were often embedded within other Most current state-of-the-art models deal problem embedded/nested entity very complex neural network architectures. In this work, we proposed to solve nested named-entity using transfer-learning approach. For purpose, different variants fine-tuned, pretrained, BERT-based used for joint-labeling modeling technique. Two named-entity-recognition datasets, i.e., GENIA and GermEval 2014, experiment, four two levels annotation, respectively. Also, experiments performed on JNLPBA dataset, which has annotation. performance above measured F1-score metrics, commonly standard metrics evaluate models. addition, approach compared conditional random field Bi-LSTM-CRF model. It found outperformed significantly without requiring any external resources or feature extraction. results existing approaches. best-performing model achieved F1-scores 74.38, 85.29, 80.68 GENIA, transfer learning (i.e., pretrained BERT after fine-tuning) based task could perform well a more generalized comparison many