Skip to content

Dataset and codes for our IJCAI 2019 paper "Adapting BERT for Target-Oriented Multimodal Sentiment Classification"

Notifications You must be signed in to change notification settings

jefferyYu/TomBERT

Repository files navigation

TomBERT

Adapting BERT for Target-Oriented Multimodal Sentiment Classification

Author

Jianfei YU

jfyu@njust.edu.cn

Mar 02, 2020

Target-oriented Multimodal Sentiment Classification (TMSC), PyTorch Implementations.

Requirement

  • PyTorch 1.0.0
  • Python 3.7

Download tweet images and set up image path

Code Usage

(Optional) Preprocessing

  • This is optional, because I have provided the pre-processed data under the folder named "absa_data"
python process_absa_data.py

Training for TomBERT

  • This is the training code of tuning parameters on the dev set, and testing on the test set. Note that you can change "CUDA_VISIBLE_DEVICES=6" based on your available GPUs.
sh run_multimodal_classifier.sh

Testing for TomBERT

  • After training the model, the following code is used for directly loading the trained model and testing it on the test set
sh run_multimodal_classifier_test.sh

Implemented models

BERT and BERT+BL (run_classifier.py)

  • You can run the following code to perform training and testing.
sh run_classifier.sh

TomBERT, mBERT, Res-BERT (run_multimodal_classifier.py)

  • You can choose different models in the "run_multimodal_classifier.sh" file.

BERT and TomBERT trained by me

Acknowledgements

About

Dataset and codes for our IJCAI 2019 paper "Adapting BERT for Target-Oriented Multimodal Sentiment Classification"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published