Awesome Open Source
Awesome Open Source

GAN for Text Summarization

Tensorflow re-implementation of Generative Adversarial Network for Abstractive Text Summarization.

Requirements

  • Python3
  • Tensorflow >= 1.4 (tested on Tensorflow 1.4.1)
  • numpy
  • tqdm
  • sklearn
  • rouge
  • pyrouge

You can use the python package manager of your choice (pip/conda) to install the dependencies. The code is tested on Ubuntu 16.04 operating system.

Quickstart

  • Dataset

    Please follow the instructions here for downloading and preprocessing the CNN/DailyMail dataset. After that, store data files train.bin, val.bin, test.bin and vocabulary file vocab into specified data directory, e.g., ./data/.

  • Prepare negative samples for discriminator

    You can download the prepared data discriminator_train_data.npz for discriminator from dropbox and store into specified data directory, e.g., ./data/.

  • Train the full model

    python3 main.py --mode=train --data_path=./data/train.bin --vocab_path=./data/vocab --log_root=./log --pretrain_dis_data_path=./data/discriminator_train_data.npz --restore_best_model=False
    
  • Decode

    python3 main.py --mode=decode --data_path=./data/test.bin --vocab_path=./data/vocab --log_root=./log --single_pass=True
    

References


Get A Weekly Email With Trending Projects For These Topics
No Spam. Unsubscribe easily at any time.
Python (1,123,905
Tensorflow (12,537
Generative Adversarial Network (1,346
Text Summarization (217
Related Projects