Persuasive Orderings

This repo contains code for the following paper:

Omar Shaikh, Jiaao Chen, Jon Saad-Falcon, Duen Horng (Polo) Chau, Diyi Yang: Examining the Ordering of Rhetorical Strategies in Persuasive Requests (EMNLP (Findings) 2020)

To see an overview of our analyses, take a look at pattern_finding.ipynb.

Requirements

• Python 3.6 or higher
• Pytorch >= 1.3.0
• transformers
• Pandas, Numpy, Pickle

Code Structure

|__code/
        |__ vae_train/ --> folder for our VAE model. Run train.py to train this model.
        |__ dataset_iterators.py --> specific iterators for different analyses.
        |__ pattern_finding.ipynb --> step by step breakdown of the analysis in the paper.
        |__ baselines.ipynb --> notebook for evaluating baselines.
        |__ lstm_train/ --> folder for our LSTM model. Import the train method from train.py, and follow steps in pattern_finding.ipynb to train this.
        |__ editing_utils.py/ --> utilities to handle edits made to requests

Instructions

Training the VAE

Please run train.py in code/vae_train/

Analyzing + Training LSTM

Please run ./pattern_finding.ipynb

Running Baselines

Please run ./baselines.ipynb to train the BERT baseline model and Naive Bayes.