baseline_CRF

Baseline Linear CRF:

We provided a simple Liner CRF model built using contextual, lexical and gazetter features. The following script contains the baseline Feature Based CRF model:

      ./crf_ner_wlp.py

The predicted outputs of this model on the test data will be stored as both StandOff and CoNLL fomat in the following directories respectively:

      ./Standoff_Outputs/
      ./Conll_Outputs/

To run and get the prediction on the test data, you need to provide the script with the following:

1. The <location of StandOff format train data> in -train_data parameter, and
2. The <location of StandOff format test data> in the -test_data parameter

python crf_ner_wlp.py  -train_data "../../data/train_data/Standoff_Format/" -test_data "../../data/test_data/Standoff_Format/"

Optional arguments for Feature Ablation

The following arguments can be passed to the crf_ner_wlp.py for feature ablation:

  -include_word_features INCLUDE_WORD_FEATURES
                        string feature inclusion (0 to disable)
  -include_context INCLUDE_CONTEXT
                        context feature inclusion (0 to disable)
  -include_gazetteer INCLUDE_GAZETTEER
                        gazetteer feature inclusion (0 to disable) 
                        [gazetters used in this model can be found in the  'gazetters/' folder]

Example: Exclude String Features: python crf_ner.py -include_camel_case 0 -include_word_features 0

Example: Exclude Context Features: python crf_ner.py -include_context 0 -include_word_features 0

Example: Exclude Gazetteer Features: python crf_ner.py -include_gazetteer 0

Evaluting model performance:

The following script should be used to evaluate the performance of the model predictions:

  ./evalutation.py

Please see the detailed evalution process at: Eval folder