Hierarchical Attention Networks

Pytorch/Tensorflow implementation of Hierarchical Attention Networks for Document Classification.
Model has a hierarchical structure that mirrors the hierarchical structure of documents, and consist of word-level encoder/attention layer, sentence-level encoder/attention layer.

Requirements

• Pytorch or Tensorflow, nltk, NumPy, pandas, matplotlib

Data

• Sample_text.zip (Sample_text.csv)
• Data consist of 100,000 reviews and stars.

class	text
4	"It was a great experience. They helped us ... "
5	"Amazing service to use for removing junk ..."
1	"Good little cafe in Matthews. I tried the ..."
...	...

Use

1. Download sample_text.zip and unzip
2. run python prep.py for text preprocessing
3. run python pytorch_main.py or python tf_main.py for traning (check argument)
4. run python pytorch_main.py --mode='test' --test_iters=***
   or python tf_main.py --mode='test' --test_iters=*** for test

Result