Aspect-Level Sentiment Analysis Via Convolution over Dependency Tree

Dataset and code for the paper: Aspect-Level Sentiment Analysis Via Convolution over Dependency Tree. Kai Sun, Richong Zhang, Samuel Mensah, Yongyi Mao, Xudong Liu. EMNLP 2019. [pdf]

Overview

A dependency tree shortens the distance between the aspects and opinion words of a sentence, allowing neural network models to capture long-term syntactic dependencies effortlessly. Besides, dependency trees have graph-like structures bringing to play the recent class of neural networks, namely, graph convolutional networks (GCN). The GCN has been successful in learning representations for nodes, capturing the local position of nodes in the graph. These observations motivate us to develop a neural model which can operate on the dependency tree of a sentence, with the aim to make accurate sentiment predictions with respect to specific aspects. Specifically, a BiLSTM is first employed to capture the contexutual information for each word. Then the BiLSTM embeddings are enhanced by a multi-layer GCN over the dependency tree. The aspect-sepcific embeddings of the last GCN layer are extracted and used for final classification.

Requirement

• Python 3.6.7

• PyTorch 1.2.0

• NumPy 1.17.2

• GloVe pre-trained word vectors:

  • Download pre-trained word vectors here.
  • Put glove.840B.300d.txt into the dataset/glove/ folder.

• Rported results in the paper are under a fixed random seed, thus results might be unstable under different GPU devices or random seeds. To reproduce the reported results, you can try to train the model for several times under different random seeds such as from 0 to 50. If you want to get the trained models, please refer to https://drive.google.com/file/d/1ijAnzl1pHtSimRsxBEVoArVg4iJw18zg/view?usp=sharing

• Note that some sentences in Rest16 dataset have no aspects, thus in fact sentence-level. Most of recenct works on ABSA have removed these sentences for training and evaluation.

Usage

Training the model:

python train.py --dataset [dataset]

Prepare vocabulary files for the dataset:

python prepare_vocab.py --dataset [dataset]

Evaluate trained model

python eval.py --model_dir [model_file path]

Citation

If this work is helpful, please cite as:

@inproceedings{Sun2019CDT,
  author    = {Kai Sun and
               Richong Zhang and
               Samuel Mensah and
               Yongyi Mao and
               Xudong Liu},
  title     = {Aspect-Level Sentiment Analysis Via Convolution over Dependency Tree},
  booktitle = {Proceedings of the 2019 Conference on Empirical Methods in Natural
               Language Processing and the 9th International Joint Conference on
               Natural Language Processing, {EMNLP-IJCNLP} 2019, Hong Kong, China,
               November 3-7, 2019},
  pages     = {5678--5687},
  year      = {2019}
}

License

MIT