media-election-interference

What problem does it solve?

Election-safety research requires standardized benchmarks to evaluate how models handle manipulative political content. This Skill provides a realistic evaluation pipeline that generates or curates voter manipulation messages and amplification strategies across five tactics to stress-test stance and sentiment classifiers.

Core Features & Use Cases

• Five-tactic benchmark dataset with realistic messages and amplification strategies (voter_suppression, astroturfing, micro_targeting, foreign_influence, deepfake_candidate).
• Dual-model validation using stance detection and sentiment analysis to assess model outputs.
• Quality gates enforce minimum word counts and placeholder checks to ensure detailed, actionable data.
• Dataset: campaigns.csv provides platform, demographic, and reach context for evaluation.
• Use cases include evaluating model safety, robustness, and response to political persuasion scenarios in research workflows.

Quick Start

Run python benchmark.py to execute the five-tactic ISC benchmark against the campaigns.csv dataset.