Personalized Author Obfuscation with Large Language Models

2025-05-17

AI Summary

This work addresses the significant individual variability in large language model (LLM)-based author anonymization, where existing methods yield bimodal author obfuscation performance—effectively hiding some authors while failing others. To tackle this, we propose a personalized style rewriting framework featuring a style-aware rewriting mechanism and user-adaptive prompt engineering, enabling fine-grained, author-specific writing style transfer. We further introduce an author-level evaluation protocol to quantify individual obfuscation efficacy. Experiments across multiple author identification benchmarks demonstrate an average 18.7% improvement in concealment performance. Critically, the method substantially enhances obfuscation for previously low-performing authors and narrows inter-user performance disparities. This is the first work to systematically identify and mitigate the “personalized failure” problem in LLM-based author anonymization.

Technology Category

• Machine Learning: Privacy
• Natural Language Processing: (Large) Language Models
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

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.

Problem

• Evaluate LLMs for authorship obfuscation via style alteration
• Analyze user-wise obfuscation effectiveness variations
• Propose personalized prompting to mitigate efficacy bimodality

Innovation

• Uses LLMs for personalized author obfuscation
• Analyzes user-wise obfuscation effectiveness variation
• Proposes personalized prompting to mitigate bimodality