ryanzhumich / dscnn

Dependency Sensitive Convolutional Neural Networks

Code for the paper Dependency Sensitive Convolutional Neural Networks for Modeling Sentences and Documents (NAACL 2016)

Demo with TREC dataset

The implementation is based on https://github.com/yoonkim/CNN_sentence and http://deeplearning.net/tutorial/lstm.html

Dependencies

• Python (2.7)
• Theano (0.8)
• Pandas (0.17)

Prepare Pretrained Word Embeddings

The model uses preptrained word embeddings including word2vec and GloVe. Download those word embeddings and save them as:

• word2vec: data/GoogleNews-vectors-negative300.bin
• GloVe: data/glove.840B.300d.txt

Data Preprocessing

cd code/preprocess/
python process_trec.py

Run Demo with Training and Testing

cd code/
./run_trec_demo.sh