Unsupervised Detection of Ideological Bias in Contextualized Embeddings

This repository contains the code for the ICML 2022 paper Unsupervised Detection of Contextualized Embedding Bias with Application to Ideology.

Dependencies

The code requires Python>=3.6, numpy>=1.18, torch>=1.2, and torch_geometric>=1.6.

Usage

To replicate the experiments on finding the ideological subspace, run the script src/model/train.sh. To replicate the experiments without rotation and sparsity, run the script src/model/train_ra_sa.sh.

The scripts expect pickled year-specific datasets in data/final/, which can be created using src/data/prepare_data.sh.

Citation

If you use the code in this repository, please cite the following paper:

@inproceedings{hofmann2022unsupervised,
    title = {Unsupervised Detection of Contextualized Embedding Bias with Application to Ideology},
    author = {Hofmann, Valentin and Pierrehumbert, Janet and Sch{\"u}tze, Hinrich},
    booktitle = {Proceedings of the 39th International Conference on Machine Learning},
    year = {2022}
}