Personalized Author Obfuscation with Large Language Models

TL;DR

This paper explores using large language models to obfuscate authorship by paraphrasing, revealing that personalized prompting improves effectiveness and reduces variability across individual users.

Contribution

It introduces a personalized prompting technique for LLM-based author obfuscation, addressing variability in effectiveness across users.

Findings

LLMs are generally effective at author obfuscation.

Performance varies significantly across individual users.

Personalized prompting improves obfuscation success and reduces bimodal efficacy distribution.

Abstract

In this paper, we investigate the efficacy of large language models (LLMs) in obfuscating authorship by paraphrasing and altering writing styles. Rather than adopting a holistic approach that evaluates performance across the entire dataset, we focus on user-wise performance to analyze how obfuscation effectiveness varies across individual authors. While LLMs are generally effective, we observe a bimodal distribution of efficacy, with performance varying significantly across users. To address this, we propose a personalized prompting method that outperforms standard prompting techniques and partially mitigates the bimodality issue.