PURE name entity recognition model for ace2005 Chinese
a name entity recognition model base on PURE for Chinese ace2005 dataset. Comparing with origin PURE, I just removed allennlp package dependence.
step1: run ace_parser.py
to parse ace2005 data, the code is based on ace2005chinese_preprocess,I have modified the code to adapte PURE model.
step2: run train.sh
to train model.
Result on hfl/chinese-bert-wwm-ext with 21 entity types including event-type:
INFO - root - P: 0.83622, R: 0.63672, F1: 0.72296
notice that PURE model evaluate the model useing a strict standard which means a correct result must have a correct start_index, end_index and entity_type, and the result is phrase level not tag level.