Implementation of document classification model described in Hierarchical Attention Networks for Document Classification (Yang et al., 2016).
Get Yelp review dataset here: https://www.yelp.com/dataset_challenge
python3 yelp_prepare.py yelp_academic_dataset_review.json
python3 worker.py --mode=train --device=/gpu:0 --batch-size=30
I am getting 65% accuracy on a dev set (16% of data) after 3 epochs. Results reported in the paper are 71% on Yelp'15. No systemic hyperparameter optimization was performed.