2021 Social Network and Public Opinion Processing

Paper recurrence: EMNLP2016 paper Keyphrase Extraction Using Deep Recurrent Neural Networks on Twitter

Rewritten on the basis of Source codes by PyTorch implementation & Modified tensorflow implementation

You may need to prepare the pre-trained word vectors (or not).

• Pre-trained word vectors. Download GoogleNews-vectors-negative300.bin.gz

Joint RNN model

• data: Dataset

• checkpoints: Store the parameters obtained from model training

• main.py: Main program (tensorflow implementation)

• main_v2.py: Main program (PyTorch with 2 separate RNN layers)

• main_v2_cell.py: Main program (PyTorch with 2-layer RNNCell）

• model.py: Defined the joint-rnn model (tensorflow)

• model_v2.py: Defined the joint-rnn model (PyTorch)

• bi_lstm_model.py: Use bidirectional lstm instead of rnn (part of the code has not been modified, if you want to run, please modify it according to the format of model.py first)

• load.py: Load the dataset

• tools.py: Some utility functions

• tensorflow 2.2.0 + tensorlayer
• PyTorch

python data/data_process.py
python main.py  // original tensorflow
python main_v2.py  // PyTorch