A Survey of Low-Resource Named Entity Recognition

Named Entity Recognition, as one of the typical tasks of information extraction, has a wide range of applications. However, the difficulty of data collection and the lack of data annotation are common problems in real scenarios. It is difficult for classical named entity recognition methods to fully obtain hidden information when the dataset and data labels are insufficient. In this case, the recognition accuracy will drop significantly. In resource-poor scenarios, the use of multi-task learning, data augmentation, and transfer learning methods can effectively utilize limited data resources or introduce data from other fields to optimize model performance. In this survey, firstly, we introduce the reader to the state of research on named entity recognition, and outline the reasons for the survey and the contribution of this paper. Then, we comprehensively describe the datasets and evaluation methods included in named entity recognition, and propose a Classification of named entity recognition in the case of lack of resources. Finally, we analyze the existing problems on the basis of the above, and further look forward to the future development direction.