This is the base code for Dynamics of Polarizing Rhetoric in Congressional Tweets

1. install pytorch (CUDA recommended if GPU available)
2. clone this git and install requirements
3. install mongoDB, set up a database named wanderingpole and a collection named tweets

Using the new academic license for Twitter, it is possible to collect a full collection of tweets for each user. This is done using collect.py in the collectingtweets directory.

Relies on the terrific HuggingFace versions of RoBERTa/BERTweet. Run with train_classifier.py.

The latest (pretrained) model we use for classifying polarization is available on huggingface

After training a model (or downloading the pre-trained model), classification can be run on the tweets in the mongoDB collection using the files in the classify_tweets directory

Various scripts available in the analyses directory.