Implementation of a seq2seq model for summarization of textual data using the latest version of tensorflow.
Demonstrated on amazon reviews, github issues and news articles.
I tried the network on three different datasets.
- Amazon Fine Food Reviews dataset
- Github issues dataset
- All the news dataset
All three of them are available on Kaggle:
I uploaded three .py and three .ipynb files. The .py files contain the network implementation and utilities. The notebooks are demos of how to apply the model. Maybe it is useful for someone.
- Bidirectional RNN
- Bahdanau Attention
- Adam Optimizer
- exponential or cyclic learning rate
- Beam Search or Greedy Decoding