gary083/sentiment_transfer_network

sentiment_transfer_network

Generative Adversarial Network based：

Which is based on Style Transfer from Non-Parallel Text by Cross-Alignment

Prerequisties:

• python 3.6 or higher
• tensroflow 1.3 or higher
• GPU memory 6G or higher (GeForce GTX 1060 up)

usage:

1. Prepare dataset：

• Step 1 : Make a directory for your dataset.
  mkdir -r data/[your_dataset_name]
• Step 2 : Prepare training data.
  Put positive and negative datasets into the directory and rename them as pos_file.txt and neg_file.txt respectively.
  (Every sentences in pos_file.txt and neg_file.txt are split by \n)
• Step 3 : Prepare testing data.
  Put the testing data into the directory and rename it as seq2seq.txt.
  (format of seq2seq.txt is the same as pos_file.txt and neg_file.txt)

2. Training：
   

• run python main.py -train -attention -model [your_model_name] -data_path [your_dataset_name]

3. Testing：
   

• run python main.py -test -attention -model [your_model_name] -data_path [your_dataset_name]

Transfer Network based：

Prerequisties:

• python 3.6 or higher
• tensroflow 1.0

usage:

1. Prepare dataset：

• Step 1 : Make a directory for your model.
  mkdir -r works/[your_model_name]

• Step 2 : Prepare sentiment data.
  Put positive and negative datasets into the directory and rename them as pos_file.txt and neg_file.txt respectively.
  (Every sentences in pos_file.txt and neg_file.txt are split by \n)

• Step 3 : Prepare dialogue data.
  Put dialogue dataset into the directory and rename them as chat.txt.
  (Each pair of dialogue (question and answer) is split by \n)

• Step 4 : Prepare testing data.
  Put the testing data into the root directory and rename it as seq2seq.txt.
  (format of seq2seq.txt is the same as pos_file.txt and neg_file.txt)

2. Training：

• Step 1： Training the variational autoencoder.
  -- run python main.py -step1 -model [your_model_name]
• Step 2： Training the sentiment classifier.
  -- run python main.py -step2 -model [your_model_name]
• Step 3： Training the transfer network.
  -- run python main.py -step3 -model [your_model_name]

3. Testing：

• run python main.py --test -model [your_model_name]