HKUST-KnowComp / DMSC

This repository is for the paper "Document-Level Multi-Aspect Sentiment Classification as Machine Comprehension"

DMSC_MC

This repository is for our EMNLP17 paper "Document-Level Multi-Aspect Sentiment Classification as Machine Comprehension". We thank Tao Lei as our code is developed based on his code.

Setup

git clone [e-mail]:HKUST-KnowComp/DMSCMC.git
cd src
export PYTHONPATH=/path/to/DMSCMC/src

Usage

• Run experiments

cd src
THEANO_FLAGS="floatX=float32,device=gpu0,cuda.root=/usr/local/cuda,on_unused_input=ignore,optimizer=fast_compile"  python DMSCMC.py --emb  ../data/tripadvisor/embs  --train  ../data/tripadvisor/train  --dev  ../data/tripadvisor/dev --test  ../data/tripadvisor/test --save ../model/base  --aspect_seeds ../data/tripadvisor/aspect.10.words 

• Run pre-trained model

cd src
THEANO_FLAGS="floatX=float32,device=gpu0,cuda.root=/usr/local/cuda,on_unused_input=ignore,optimizer=fast_compile"  python DMSCMC.py --emb  ../data/tripadvisor/embs  --train  ../data/tripadvisor/train  --dev  ../data/tripadvisor/dev --test  ../data/tripadvisor/test --save ../model/base  --aspect_seeds ../data/tripadvisor/aspect.10.words --model ../model/base

Environment

• Python 2.7
• CUDA
• Theano >= 0.7
• Numpy