This repository contains source code (tensorflow-version) to reproduce the results presented in the paper Leap-LSTM: Enhancing Long Short-Term Memory for Text Categorization (IJCAI 2019).
@inproceedings{Leap-LSTM,
title={Leap-LSTM: Enhancing Long Short-Term Memory for Text Categorization},
author={Ting Huang, Gehui Shen, Zhi-Hong Deng},
booktitle={IJCAI},
year={2019}
}
The architecture of Leap-LSTM:
The accuracies on text classification tasks:
To run the codes, you need to
- download datasets from the github repository LEAM
- to train the model, use a command like:
python train_classifier_yelp.py --rnn_model leap-lstm --yelp_set FULL \
--gpu 0 --if_schedule 0 --decay_start 1 --target_skip_rate 0.6 \
--keep_prob_word 1.0 --skip_reg_weight 1.0 --max_gradient_norm 1.0 \
--nb_epoches 5
The details of arguments can be found in skiplstm.py