No description, website, or topics provided.
Switch branches/tags
Nothing to show
Clone or download
Jiwei li
Jiwei li assdf
Latest commit 2cea5c0 Oct 1, 2015
Type Name Latest commit message Commit time
Failed to load latest commit information.
Standard_LSTM change Sep 11, 2015
hier_LSTM change Sep 11, 2015
hier_LSTM_Attention assdf Sep 30, 2015
misc standard Jun 7, 2015 asfsa Jun 20, 2015
a.txt asdf Sep 25, 2015
index.html asdf Sep 25, 2015

A Hierarchical Neural Autoencoder for Paragraphs and Documents

Implementations of the three models presented in the paper "A Hierarchical Neural Autoencoder for Paragraphs and Documents" by Jiwei Li, Minh-Thang Luong and Dan Jurafsky, ACL 2015



matlab >= 2014b

memory >= 4GB


Standard_LSTM: Standard LSTM Autoencoder

hier_LSTM: Hierarchical LSTM Autoencoder

hier_LSTM_Attention: Hierarchical LSTM Autoencoder with Attention

DownLoad Data

  • dictionary: vocabulary
  • train_permute.txt: training data for standard Model. Each line corresponds to one document/paragraph
  • train_source_permute_segment.txt: source training data for hierarchical Models. Each line corresponds to one sentence. An empty line starts a new document/sentence. Documents are reversed.
  • test_source_permute_segment.txt: target training data for hierarchical Model.

Training roughly takes 2-3 weeks for standard models and 4-6 weeks for hierarchical models on a K40 GPU machine.

For any question or bug with the code, feel free to contact

    title={A Hierarchical Neural Autoencoder for Paragraphs and Documents},
    author={Li, Jiwei and Luong, Minh-Thang and Jurafsky, Dan},
    journal={arXiv preprint arXiv:1506.01057},