seq2seqNER
Sequence-to-Sequence Model for progressively learning new Named Entities in the text. The model is implemented with PyTorch in Python3.
Model
The sequence-to-Sequence model consist of an encoder (Bidirectional LSTM) and a decoder (LSTM) with attention machenism implemented.
Data Format
Existing models are trained on CoNLL 2003 NER dataset. Other dataset can be used for training in the following format:
sentence_sequence<TAB>label_sequence
For example:
All passengers freed from Sudanese hijack plane . O O O O S-MISC O O O
Usage
For training:
python3 main.py --parameters_filepath=./parameters-train.ini
Hyperparameters are set in parameters-train.ini