Differentiated Attentive Representation Learning for Sentence Classification

Implementation of the paper Differentiated Attentive Representation Learning for Sentence Classification.

Environment

Tested on Python 2.7 and PyTorch 0.2.0.

Tensorboard-pytorch is needed if you want visualize the loss or accuracy of the model.

To run the DARLM, please first download some raw data (corpus & wordvec), then put them in directory data/.

TREC dataset

preprocessing

Use the following command to generate the data for model (or you can use above processed data):

python data/trec_data/preprocess.py

training

To run the DARLM on TREC dataset, use:

python train.py -config_file trec/hyper-param-trec.conf

please specify which data file you will use in the config file.

SST dataset

Refer to TREC.

Citation

If using this code, please cite:

Qianrong Zhou, Xiaojie Wang, Xuan Dong, Differentiated Attentive Representation Learning for Sentence Classification

@inproceedings{zhou2018differentiated,
  title     = {Differentiated Attentive Representation Learning for Sentence Classification},
  author    = {Qianrong Zhou and Xiaojie Wang and Xuan Dong},
  booktitle = {Proceedings of the Twenty-Sixth International Joint Conference on
               Artificial Intelligence, {IJCAI-18}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},             
  pages     = {4630--4636},
  year      = {2018},
  month     = {7},
  doi       = {10.24963/ijcai.2018/644},
  url       = {https://doi.org/10.24963/ijcai.2018/644},
}