Repository files navigation

Implementations of saliency models described in "Visualizing and Understanding Neural Models in NLP" by Jiwei Li, Xinlei Chen, Eduard Hovy and Dan Jurafsky.

GPU

Torch (nn,cutorch,cunn,nngraph)

pythonmatplotlib library (only for matrix plotting purposes)

downloaddata

Run the first-derivative saliency model:

sh saliency_derivative.sh

The saliency matrix will be stored in the file "matrix".

Run the variance saliency model:

sh saliency_variance.sh

The saliency matrix will be stored in the file "matrix".

##Folders and Filesinput.txt: the input sentence.

sentiment_bidi: training bi-directional lstms on the Stanford Sentiment Treebank. You can either download a pretrained model (sentiment_bidi/model) or train it yourself by running sentiment_bidi/main.lua

sentiment_uni: training uni-directional standard recurrent models.

data/dict.txt: word dictionary. Current models only support tokens found in the dictionary. Will fix it soon.

For any pertinent questions, feel free to contactjiweil@stanford.edu

Yoon Kim's seq2seq-attn repo

Wojciech Zaremba's lstm repo

Socher et al.,'s Stanford Sentiment Treebank dataset

@article{li2015visualizing,    title={Visualizing and understanding neural models in NLP},    author={Li, Jiwei and Chen, Xinlei and Hovy, Eduard and Jurafsky, Dan},    journal={arXiv preprint arXiv:1506.01066},    year={2015}}