To reproduce the results, all the folders and files above should be in the same directory
In the data folder we have the three datasets, Pstance, Covid 19 and Vast. The new dataset COVID CQ is also in the data/covid19-stance as hcq_train, hcq_test and hcq_val translated_datasets contains all the translated datasets for VAST and Covid 19 tweets data.
To get the result for PStance, target-specific stance detection, Biden
python run_pstance_biden.py
To get the result for PStance, target-specific stance detection, Sanders
python run_pstance_sanders.py
To get the result for PStance, target-specific stance detection, Trump
python run_pstance_trump.py
To get the result for PStance, cross-target stance detection, Biden
python run_pstance_biden2sanders.py
To get the result for COVID19-Stance, target-specific stance detection, face mask
python run_covid_mask.py
To get the result for COVID19-Stance, target-specific stance detection, HCQ
python run_covid_hcq.py
To get the result for VAST, zero/few-shot stance detection
python run_vast.py
To run XLM R, we have modified the codes and created a src_xlm_r folder. To get the result for school closures Covid data on XLM R
python run_covid_school_xlm_r.py
To get the result for stay at home Covid data on XLM R
python run_covid_home_xlm_r.py
To run on translated_datasets, first download the respective train,test and val files from translated_datasets folder and replace the existing train,test and val files in data folder. For example, to run VAST on German data, first download the train,test and dev files from german folder in translated_datasets folder and replace them with train,test and dev files in data folder
To perform error analysis, we have modified the Engine.py file and created a src_error_analysis folder. We performed the error analysis for PStance, target-specific stance detection, Biden. To get the error analysis results, run the following command:
python run_err_biden.py