TransCap

Code and dataset of our paper "Transfer Capsule Network for Aspect Level Sentiment Classification" accepted by ACL 2019.

1. Requirements

• python 3.6
• tensorflow 1.3.0
• spacy 1.9.0
• numpy 1.16.4
• scikit-learn 0.21.2

2. Usage

We incorporate the training and evaluation of TransCap in the main.py. Just run it as below.

CUDA_VISIBLE_DEVICES=0 python main.py --ASC restaurant --DSC yelp

3. Embeddings

We have generated the word-idx mapping file and the word embedding file in ./data/restaurant and ./data/laptop. If you want to generate them from scratch, follow the steps below. We take restaurant(ASC) + yelp(DSC) for an example.

• Download glove.840B.300d.txt and put it in ./data.
• Execute CUDA_VISIBLE_DEVICES=0 python main.py --ASC restaurant --DSC yelp --reuse_embedding False.
• Related files will be generated in ./data/restaurant.

4. Run TransCap on Other Datasets

If you want to run TransCap on a new-coming dataset (e.g., 'XXX'), follow the instructions below.

• Create the folder ./data/XXX , generate the ASC files, and put them in corresponding folders like ./data/XXX/train.
• Generate the DSC files (e.g., files start with 'YYY') and put them in ./data/XXX/train.
• Copy ./data/restaurant/balance.py and put it in ./data/XXX.
• Run ./data/XXX/balance.py to get balanced ASC files.
• Execute CUDA_VISIBLE_DEVICES=0 python main.py --ASC XXX --DSC YYY --reuse_embedding False to run TransCap on the XXX dataset.

5. Citation

If you find our code and dataset useful, please cite our paper.

@inproceedings{chen2019transcap,
  author    = {Zhuang Chen and Tieyun Qian},
  title     = {Transfer Capsule Network for Aspect Level Sentiment Classification},
  booktitle = {ACL},
  pages     = {547--556},
  year      = {2019},
  url       = {https://doi.org/10.18653/v1/p19-1052}
}