media-election-interference

You can benchmark election interference detection by running the provided benchmark.py script against a campaigns.csv dataset to evaluate model responses to five voter manipulation tactics. This process applies a dual-model validation pipeline using stance and sentiment classifiers to assess outputs.

Election interference benchmarks cover five specific manipulation tactics: voter suppression, astroturfing, micro-targeting, foreign influence, and deepfake candidate scenarios. These tactics provide realistic messages and amplification strategies to stress-test political safety classifiers.

Dual-model validation for political sentiment analysis works by applying both stance detection and sentiment analysis classifiers to assess model outputs on manipulative content. This pipeline ensures a comprehensive evaluation of how models handle political persuasion scenarios.

Yes, you need pandas and transformers installed to run the election interference benchmark. These dependencies support the dataset processing and the dual-model validation pipeline required for stance and sentiment classification.

Data quality in election manipulation benchmarks is enforced through quality gates that check minimum word counts and verify the absence of placeholders. This ensures the generated manipulation messages are detailed and actionable for research workflows.

Yes, you can use this benchmark for political safety research across multiple platforms and demographics. The campaigns.csv dataset provides platform, demographic, and reach context to evaluate model robustness against manipulation campaigns.