Notebook for the detection of attack/support relation via Adversarial Training
- Do the finetuning with the whole dataset for Debate
- Create the model that injects discourse markers into the input text and then finetunes [DONE]
- Find other candidate datasets to check [DONE: m-arg]
- Run the tests on the additional candidate datasets
- Try more comprehensive tests: apply a gridsearch for the loss weights; [DONE]
- try to add more data for discovery with proper class weighting [DONE]
- Try more sophisticated architectures. Specifically:
- Try adding an additional GRL layer on the predictions;
- Try testing other adversarial methodologies
- Research on other architectures
- Apply various analysis methodologies