Style-Aware Paraphrasing for Reducing Authorship Attribution

This LLM-driven approach constructs stylistic profiles to rewrite text, suppressing identifiable fingerprints while preserving semantic meaning. The method reduces authorship attribution F1 scores by 60-70% across blog and review datasets without the utility loss seen in differential privacy methods.