The success of neural network based methods in named entity recognition (NER) is heavily relied on abundant manual labeled data. However, these NER methods are unavailable when the data is fully-unlabeled in a new domain. To address the problem, we propose an unsupervised cross-domain model which leverages labeled data from source domain to predict entities in unlabeled target domain. To relieve the distribution divergence when transferring knowledge from source to target domain, we apply adversarial training. Furthermore, we design an entity-aware attention module to guide the adversarial training to reduce the discrepancy of entity features between different domains. Experimental results demonstrate that our model outperforms other methods and achieves state-of-the-art performance.
Bibliographical noteFunding Information:
An abstract conference version of this article has been published at ( Peng et al., 2021 ). Compared to the abstract version, this article has been significantly extended in terms of the literature review, proposed model, experiments, and result analysis. This work was supported by the National Natural Science Foundation of China (No. 62076100 ), National Key Research and Development Program of China (Standard knowledge graph for epidemic prevention and production recovering intelligent service platform and its applications) , the Fundamental Research Funds for the Central Universities , SCUT (No. 2017ZD048 , D2182480 ), the Science and Technology Planning Project of Guangdong Province (No. 2017B050506004 ), the Science and Technology Programs of Guangzhou (No. 201704030076 , 201802010027 , 201902010046 ) and the Hong Kong Research Grants Council (project no. C1031-18G ).
- Adversarial training
- Entity-aware attention
- Named entity recognition
- Unsupervised cross-domain