D3NER NER System

A D3NER NER System is a BiLSTM-CRF Training System that is used to solves a Named Entity Recognition Task.

• Context:
  • It can be used for:
    • Running D3NER main program:

      python main.pyc [-h] model dataset input_file output_file

    • Evaluating a pre-trained model:

      python -m train.evaluate [-h] [-cf] model dataset test_set

    • Building data for model training and evaluation:

      python -m train.build_data [-h] dataset train_set dev_set test_set word_embedding ab3p;

    • Training new model:

      python -m train.run [-h] [-es | -e EPOCH] [-v] model dataset train_set dev_set

where model is the name of the model being used; dataset the name of the dataset that the model was trained on; input_file is path to the input file; output_file path to the output file; train_set is path to the training dataset; dev_set is the path to the development dataset; test_set is the path to the test dataset; word_embedding is path to the word embedding pre-trained model (e.g. wikipedia-pubmed-and-PMC-w2v.bin); ab3p path to the Ab3P program. -h shows help message; -cf prints out the confusion_matrix; -es performs an early stop; -e EPOCH, prints out the number of epochs to train; -v prints ouy training process.

• Example(s):
  • Running a CDR test file with a trained model on CDR corpus:

    python main.pyc d3ner_cdr cdr data/cdr/cdr_test.txt output.txt;

  • Evaluating the model trained on CDR corpus using CDR test data and also report the confusion matrix:

    python -m train.evaluate d3ner_cdr cdr data/cdr/cdr_test.txt -cf

  • Training new model on CDR corpus with early stopping option:

    python -m train.run d3ner_cdr cdr data/cdr/cdr_train.txt data/cdr/cdr_dev.txt -es.

  • …
• Counter-Example(s):
  • a Att-BiLSTM-CRF Training System.
  • a Bidirectional LSTM-CNN-CRF Training System,
  • an Unidirectional LSTM-based Language Modeling System.
  • an Unidirectional LSTM Recurrent Neural Network Training System.
  • a seq2seq-based Neural Modeling System.
  • a Bidirectional LSTM-CNN Training System.
  • a Deep Stacked Bidirectional LSTM Recurrent Neural Network Training System.
  • a Deep Stacked Unidirectional LSTM Recurrent Neural Network Training System.
• See: Abbreviation Plus Pseudo-Precision (Ab3P), LSTM Training System, neuroner.com, Conditional Random Field, Bidirectional Recurrent Neural Network.

---

---

References

2018a

• (Github, 2018) ⇒ AiDante-D3NER: https://github.com/aidantee/D3NER Retrieved: 2018-07-01
  • QUOTE: D3NER, version 1.0, is a program that was developed by AiDante team. The program has 3 main purposes:
    • Recognizing disease and chemical entities in text documents,
    • Evaluating pre-trained models with test dataset,
    • Training new models with given corpora that follow the BioCreative V format.

2018b

• (Le et al., 2018) ⇒ Hoang-Quynh Le, Trang M Nguyen, Sinh T Vu, Thanh Hai Dang (2018). D3NER: Biomedical named entity recognition using CRF-biLSTM improved with fine-tuned embeddings of various linguistic information. Bioinformatics.
  • ABSTRACT: Results We propose D3NER, a novel biomedical named entity recognition (NER) using conditional random fields and bidirectional long short-term memory improved with fine-tuned embeddings of various linguistic information. D3NER is thoroughly compared with seven very recent state-of-the-art NER models, of which two are even joint models with named entity normalization (NEN), which was proven to bring performance improvements to NER. Experimental results on benchmark datasets, i.e. the BioCreative V Chemical Disease Relation (BC5 CDR), the NCBI Disease, and the FSU-PRGE gene/protein corpus, demonstrate the out-performance and stability of D3NER over all compared models for chemical, gene/protein NER and over all models (without NEN jointed, as D3NER) for disease NER, in almost all cases. On the BC5 CDR corpus, D3NER achieves F1 of 93:14% and 84:68% for the chemical and disease NER, respectively; while on the NCBI Disease corpus, its F1 for the disease NER is 84:41%. Its F1 for the gene/protein NER on FSU-PRGE is 87:62%.

---