Neural-based models in NLG

Experiments using neural network RNN(char and word level) and GAN(char-level) in Natural Language Generation and their performance

raw_corpus/

The Big Bang Theory S1-S10: https://github.com/skashyap7/TBBTCorpus/tree/master/preprocessing

Crawl from here: https://bigbangtrans.wordpress.com/

data/

Training data used for the experiment. Extracted all Sheldon transcripts from raw_corpus

doc/

All related report and presentation

preprocess/

Codes for crawling data and preprocessing

rnn_based_solution/

Codes for char and word level generation using RNN

GAN_result/

Results of sample output and training logs after training GAN in char level with sequence length of 80, codes are in the reference

Reference

1. RNN model:

• https://github.com/martin-gorner/tensorflow-rnn-shakespeare

• https://github.com/jamesrequa/TV-Script-RNN

• https://github.com/shawnwun/RNNLG

2. GAN:

• https://github.com/amirbar/rnn.wgan

3. Crawler:

• https://github.com/skashyap7/TBBTCorpus/tree/master/preprocessing