GitHub - jialuli-luka/Debate_Persuation_Argument_Structure (original) (raw)

Exploring the Role of Argument Structure in Online Debate Persuasion

Code for paper "Exploring the Role of Argument Structure in Online Debate Persuasion" (https://www.aclweb.org/anthology/2020.emnlp-main.716.pdf) in EMNLP 2020.
Data preparation: Download DDO data from https://www.cs.cornell.edu/~esindurmus/ddo.html

Generate Argument Structure follow https://github.com/vene/marseille (use SVM & strict option) save the generated argument structure for each debate under src/data/. (each debate saved as one json file)

Filter data: python src/generation_stat.py

To replicate the full model in the paper:

1. Filter out data based on criterion (e.g. length, vote difference) (Change CORENLP_PATH at the beginning of the file) python src/BERT/data_generation.py
2. Generate BERT representation for debates python src/BERT/BERT_representation_generation.py
3. Train full model: python src/BERT/RNN_with_BERT_experiment.py full
4. Train model w/o Argument Structure: python src/BERT/RNN_with_BERT_experiment.py linguistic

To replicate LR baselines in the paper: python src/experiment.py