Many named entities are embedded in others, but current models just only focus on recognizing entities at the top-level. In this paper, we proposed two approaches for the nested named entity recognition task by modeling this task as the multilayer sequence labeling task. Firstly, we propose a model that integrates linguistic features with a neural network to improve the performance of named entity recognition (NER) systems, then we recognize nested named entities by using a sequence of those models in which each model is responsible for predicting named entities at each layer. This approach seems to be inconvenient because we need to train many single models to predict nested named entities. In the second approach, we proposed a novel model, called multilayer recurrent neural networks, to recognize all nested entities at the same time. Experimental results on the Vietnamese data set show that the proposed models outperform previous approaches. Our model yields the state of the art results for Vietnamese with F1 scores of 92.97% at top-level and 74.74% at the nested level. For English, our NER systems also produce better performance.
CITATION STYLE
Nguyen, T. S., & Nguyen, L. M. (2018). Nested Named Entity Recognition Using Multilayer Recurrent Neural Networks. In Communications in Computer and Information Science (Vol. 781, pp. 233–246). Springer Verlag. https://doi.org/10.1007/978-981-10-8438-6_19
Mendeley helps you to discover research relevant for your work.