Name Entity Recognition for Military Based on Domain Adaptive Embedding

In order to solve the problem of poor quality of domain embedding space caused by insufficient corpus in a single military domain, which makes the deep learning neural network model to identify military named entities with low accuracy, this paper starts with the distributed representation of words, and uses the domain adaptation method from additional domains. Introduce more useful information to help learn the embedding in the military domain. First, establish a domain dictionary, combine it with the CRF algorithm, and perform domain-adaptive word segmentation on the collected general domain corpus and military domain corpus as the embedding training corpus, and the word vector As a feature and word vector splicing, to enrich the embedding information and verify the word segmentation effect; then perform domain adaptive transformation on the heterogeneous embedding space in the general field and military field obtained from training, generate domain adaptive embedding, and use it as the basic model BiLSTM-C