This is the code for the paper, Written Justifications are Key to Aggregate Crowdsourced Forecasts.
To install dependencies, pip install the libraries listed in requirements.txt.
Model.py contains 2 model architectures implemented using HuggingFace: The LSTM that did not concatenate question information (LSTM_Model), and the LSTM that did concatenate question information (LSTM_Model_With_Question)
Utils.py contains the code for processing the actual GJO Questions (which are found in data/), and creating the train/dev/test splits. The actual train/dev/test splits I used are found in questions.save.
Train.py contains the code for initializing all the hyperparameters of the model and the code for the training and testing loops/printing out the results.
To replicate, run utils.py then train.py.