Bias Detection in Political Social Media

This repository contains code and notebooks for the project Bias Detection in Political Social Media, a cross-platform study of political bias in short text using data from Truth Social and Bluesky, with additional experiments on U.S. Congressional tweets (2008–2017) and the Politics.com (2005) dataset.

The project builds an LLM-annotated corpus (~130k posts) with Left / Neutral / Right labels, benchmarks Naive Bayes and DistilBERT classifiers, and analyzes cross-platform semantics and temporal / domain shift.

Note: Raw datasets are not included in this repository. The main labeled corpus is available on Hugging Face: averrie/bluesky-truthsocial-political-stances

Citing

If you use this code or dataset in academic work, please cite:

@misc{huang2025bias-political-social-media,
  title        = {Bias Detection in Political Social Media},
  author       = {Huang, Xingrui and Xu, Liangyou},
  year         = {2025},
  howpublished = {\url{https://github.com/averrie/cs7980}},
}