VistaNet

This is the code for the paper:

VistaNet: Visual Aspect Attention Network for Multimodal Sentiment Analysis
Quoc-Tuan Truong and Hady W. Lauw
Presented at AAAI 2019

We provide:

• Code to train and evaluate the model
• Data used for the experiments

If you find the code and data useful in your research, please cite:

@inproceedings{truong2019vistanet,
  title={VistaNet: Visual Aspect Attention Network for Multimodal Sentiment Analysis},
  author={Truong, Quoc-Tuan and Lauw, Hady W},
  publisher={AAAI Press},
  year={2019},
}

Requirements

• Python 3
• Tensorflow >=1.12,<2.0
• Tqdm
• GloVe word embeddings

How to run

1. Make sure data is ready. Run script to pre-process the data:

python data_preprocess.py

2. Train VistaNet:

python train.py --hidden_dim 50 --att_dim 100 --num_images 3 --batch_size 32 --learning_rate 0.001 --num_epochs 20

Contact

Questions and discussion are welcome: www.qttruong.info