/rasa-nlp-architect

采用nlp-architect实现rasa-nlu中文意图提取和槽填充

Primary LanguagePython

介绍

这个repository主要是实现了intel的nlp-architect的intent和slot的提取，两个model

  • mtl
  • seq2seq

然后配合提供一个rasa-nlu的server，配合rasa-core使用。

Train mtl model

python train_mtl_model.py --dataset_path rasa_data/rasa_nlu_data/ -b 100 -e 10

Train seq2seq model

python train_seq2seq_model.py --dataset_path rasa_data/rasa_nlu_data/ -b 100 -e 10

Interactive

python interactive.py --model_path models/mtl/model.h5 --dataset_path rasa_data/rasa_nlu_data/

or

python interactive.py --model_path models/seq2seq/model.h5 --dataset_path rasa_data/rasa_nlu_data/

Provide a server

python server.py -m models/mtl/model.h5 -i models/mtl/model_info.dat

or

python server.py -m models/seq2seq/model.h5 -i models/seq2seq/model_info.dat

TODO

现阶段Intent只返回了概率最大的，后续需要做intent ranking.