conda create --name judge python=3.9.16
conda activate judge
pip install -r requirements.txt
For Author Profiling (AP) on Public Reddit, please download synthetic_dataset.jsonl to the folder './ap' from this link.
For Empathetic Conversation (EC), please download Track 3 and Track 4 data from link and then artciles from link. And put these 3 CSV under './ec'.
For OpinionQA, please download the data to './opinions_qa'.
Then, update your API Keys in utils.py.
python prism.py --num_sample 1000 --prompt_type with_persona --persona_features all_features --model gpt-4o
python opinionqa.py --num_sample 200 --prompt_type with_persona --model gpt-4o
python ap.py --model gpt-4o
python ec.py --num_article 300 --num_pair_per_article 5 --prompt_type all_features --model gpt-4o
--prompt_type controls the type of prompt to be used: with_persona, no_persona, and with_persona_with_tie
--persona_features controls the number of personas to be used: all_features, key_features, least_imp_feature
More sample experiment can be found in exp/exp.sh
All the results can be found under \outputs. The jupyter notebook to visualize the results can be found in visualization.ipynb.