Tribrid

Code Structure

The code of the model is based on the Hugging Face library of pretrained BERT. The structure is as follows:

• run_classifier.py: The utility set of the model, containing a set of dedicated self-created classess and methods that are used in training and testing the model. It is based on the pretrained BERT from huggingface.

• Tribrid.py: This file contains our proposed model for stance classificarion based on BERT with triplet (siamese) structure and a joint loss function with 3 losses.

• Tribrid_pos.py: This file contains our proposed model for stance classificarion based on BERT with siamese structure and a joint loss function with 2 losses.

• Post_Process: This folder contains the code to post process the logtis and distance for the final stance classification as well as the flyingsquid implementation.

• requirements.txt: Dependencies to run the code.

The datasets and the models are stored here:

https://drive.google.com/drive/folders/14Er5Fzy9HaYwv3Hx-bLytlwsm8VHjxYy

How to run the code

First fill in the model, dataset and output path in Tribrid.py/Tribrid_pos.py.

To run locally with a GPU :

python -u Trbrid.py/Tribrid_pos.py

Or to run using a cluster that supports prun and GPU :

prun -np 1 -native '-C TitanX --gres=gpu:1' python Trbrid.py/Tribrid_pos.py