• Use part of the re3d data set from the British Defense Science and Technology Laboratory to perform Name Entity Recognition.
• Perform word embedding processing on the preprocessed data. Combining syntactic and semantic features, try to use Word2vec and fastText for text classification, and use part-of-speech tagging, relying on parsing methods and multiple corpora to test the best method of feature embedding.
• In terms of models, use Bi-LSTM to try to combine a variety of Attention and CRF methods for training, and select the final model through the F1 score of accuracy.