/voting-booklet-bias

Code for the paper "Voting Booklet Bias: Stance Detection in Swiss Federal Communication"

Primary LanguageJupyter Notebook

Voting Booklet Bias: Stance Detection in Swiss Federal Communication

This repository contains the data and code for the paper Voting Booklet Bias: Stance Detection in Swiss Federal Communication by Eric Egli, Noah Mamié, Mathias Müller and Eyal Liron Dolev. The paper has been presented at the SwissText conference 2023 and is available here.

Data Overview

  1. Voting booklets - PDF versions of the used voting booklets (online overview)
  2. Statements - Statements extracted from the voting booklets
  3. Predictions - Predicted stances for each statement

Reproducing the Results

  1. Clone this repository:

    git clone git@github.com:ZurichNLP/voting-booklet-bias.git

  2. Init and update the submodules - this will clone the xstance repository:

    git submodule init
git submodule update

  3. Follow the instructions over in xstance to setup and train the M-BERT model*

  4. Run the predict.sh script to predict the stances for each statement:

    bash predict.sh

    Note that this will overwrite the current predictions in ./data/predictions/.

  5. (Optional) For plotting and further examining the results, you can create a new conda environment:

    conda create --name vbb2023 --file requirements.txt

    You are now ready to explore our main notebook.

* Note: We will soon provide a pre-trained model for download.

Citation

Will follow as soon as the paper is published on arXiv.

Contact

If you have any questions, do not hesitate to contact us!