Pinlong-Zhao / SDGCN

Modeling Sentiment Dependencies with Graph Convolutional Networks for Aspect-level Sentiment Classification

SDGCN

Tensorflow Implementations.

使用Tensorflow实现。

Requirement

• python 3.6 / 3.7
• tensorflow >= 1.12
• numpy
• sklearn
• bert_embedding
  • See Bert Embeddings for more detail.

Usage

Creat the embedding

• GloVe: Download pre-trained word vectors here. In this implement, we use glove.42B.300d.zip

• BERT: Refer to creat_BERT_embedding.py and creat_BERT_embedding_2_targets.py to create BERT Embedding if need.

Training

python run_glove.py 

Train model with GloVe Embedding. See run_glove.py for more training arguments.

python run_bert.py 

Train model with BERT Embedding. See run_bert.py for more training arguments.

Citation

The manuscript is avaliable in arXiv:

"Modeling Sentiment Dependencies with Graph Convolutional Networks for Aspect-level Sentiment Classification". arXiv preprint arXiv:1906.04501 (2019) [pdf]