Seq-GAN-BERT：Sequence Generative Adversarial Learning for Low-resource Name Entity Recognition

Abstract

Named entity recognition (NER), as an important basic task of natural language processing, has been widely studied. In the case of relatively sufficient labeled data, traditional NER methods have achieved remarkable results. However, due to the lack of labeled data in many fields and the difficulty of manual annotation, the task of low-resource NER has become a research hotspot. To effectively improve the recognition accuracy of low-resource NER, this paper proposes the semi-supervised learning model Seq-GAN-BERT，which integrates the adversarial generative network based on the pre-trained language model BERT, and uses the domain unlabeled corpus to train the adversarial generative network to learn the important general semantic information of the data. The proposed Seq-GAN-BERT method can further optimize BERT-based supervised training and improve the ability of entity recognition. The experimental results show that our model greatly reduces the dependence on labeled samples and effectively improves the performance of low-resource NER task.

Tasks

Reproductions