Aspect-Level Sentiment Analysis Via Convolution over Dependency Tree

Requirements

• Python 3.6.8
• PyTorch 1.0.0
• CUDA 9.0

Preparation

First, download and unzip GloVe vectors(glove.840B.300d.zip) from https://nlp.stanford.edu/projects/glove/ .

Then, put glove.840B.300d.txt into ./dataset/glove directory.

Next, prepare vocabulary with:

./build_vocab.sh

The build_vocab.sh contains commands to build vocab for different datasets.

Training

To train the CDT model, run:

./train.sh

The train.sh contains training commands for different datasets.

Model checkpoints and logs will be saved to ./saved_models.

Evaluation

You need to specify --data_dir, --vocab_dir and --model_dir in eval.py.

python eval.py