Dependency Cluster Attention for Aspect-level sentiment classification

DLCF-DCA

codes for paper Combining Dynamic Local Context Focus and Dependency Cluster Attention for Aspect-level sentiment classification. submitted to 《Neurocomputing》.

Requirement

Python >= 3.6
PyTorch >= 1.0
pytorch-transformers == 1.2.0
SpaCy >= 2.2

To use our models, you need download en_core_web_sm by python -m spacy download en_core_web_sm

Training

python train.py --model dlcf_dca

Model Architecture

dlcf_dca

Note

Some important scripts to note:

datasets/semeval14/*.seg: Preprocessed training and testing sentences in SemEval2014.
datasets/semeval15/*.seg: Preprocessed training and testing sentences in SemEval2015.
datasets/semeval16/*.seg: Preprocessed training and testing sentences in SemEval2016.
models/dlcf_dca.py: the source code of DLCF_DCA model.
data_utils.py/ABSADataSet class: preprocess the tokens and calculates the shortest distance to target words and cluster via the Dependency Syntax Parsing Tree.

Out of Memory

Since BERT models require a lot of memory. If the out-of-memory problem while training the model, here are the ways to mitigate the problem:

Reduce the training batch size ( batch_size = 4 or 8 )
Reduce the longest input sequence ( max_seq_len = 40 or 60 )
Set use_single_bert = true to use a unique BERT layer to model for both local and global contexts