Darkspace document classifier & Detection method classifier This work is developed by Xiangci Li (lixiangci8@gmail.com) on top of Dr. Gully Burns's code. Read the paper for more details.

• Tensorflow (1.9.0)
• Keras (GPU version, 2.2.2. Non-GPU version should also work.)
• h5py (2.7.1)
• Numpy (1.14.3)
• Pandas (0.23.0)
• tqdm (4.26.0)

• GloVe 50, 100, 200, 300 dimensions (http://nlp.stanford.edu/data/glove.6B.zip)
• bio_GloVe 50, 100, 300, 1024 dimensions
• FastText 300 dimensions (https://s3-us-west-1.amazonaws.com/fasttext-vectors/wiki-news-300d-1M.vec.zip)
• bio_FastText 100 dimensions

This is a document-level classifier for Darkspace dataset, whose binary label is showing whether a document mentioned interaction method or not.

To replicate the training results, run

$ cd darkspace
$ python darkspace_classification.py -n EMBEDDING_NAME -d DATA_PATH -e EMBEDDING_PATH -v VOCAB_FILE

EMBEDDING_NAME refers to the name of run files in run/EMBEDDING_NAME_run.tsv format. You can change the training settings in the run files. DATA_PATH refers to the path that the dataset is stored. EMBEDDING_PATH refers to the path that all the pretrained word embeddings are stored. VOCAB_FILE refers to the file that the vocabulary file that used when training word embeddings.

This is a paragraph-level classifier for INTACT dataset.

To replicate the training results, run

$ cd detection_method
$ python detection_method_classification.py -n EMBEDDING_NAME -d DATA_FILE -e EMBEDDING_PATH -v VOCAB_FILE

EMBEDDING_NAME refers to the name of run files in run/EMBEDDING_NAME_run.tsv format. You can change the training settings in the run files. DATA_FILE refers to the dataset file. EMBEDDING_PATH refers to the path that all the pretrained word embeddings are stored. VOCAB_FILE refers to the file that the vocabulary file that used when training word embeddings.

The only major difference in _split.py is it takes pre-splitted train, dev, test data, while the original code splits train and test during execution.

Please use the following BibTeX citation to cite our paper:

@article{burns2019building,
  title={Building deep learning models for evidence classification from the open access biomedical literature},
  author={Burns, Gully A and Li, Xiangci and Peng, Nanyun},
  journal={Database},
  volume={2019},
  year={2019},
  publisher={Oxford Academic}
}