TAROT: Task-Oriented Authorship Obfuscation Using Policy Optimization Methods

2024-07-31

arXiv.org

AI Summary

This work addresses the fundamental trade-off between author identity privacy preservation and textual utility. We propose an unsupervised author style rewriting method that, for the first time, incorporates Proximal Policy Optimization (PPO) into author obfuscation—enabling end-to-end, sentence-level rewriting via small language models without labeled data or target-author corpora. Our approach jointly optimizes privacy and utility through a multi-objective reward function and an adversarial author identification evaluation framework. Empirical results across multiple benchmarks demonstrate a >40% reduction in author identification attack accuracy, while downstream task performance (e.g., sentiment analysis, summarization) degrades by less than 2%. The method thus achieves strong privacy guarantees without compromising functional utility. Code and models are publicly released.

Technology Category

• Machine Learning: Privacy
• Natural Language Processing: Safety and Robustness
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Authorship obfuscation aims to disguise the identity of an author within a text by altering the writing style, vocabulary, syntax, and other linguistic features associated with the text author. This alteration needs to balance privacy and utility. While strong obfuscation techniques can effectively hide the author's identity, they often degrade the quality and usefulness of the text for its intended purpose. Conversely, maintaining high utility tends to provide insufficient privacy, making it easier for an adversary to de-anonymize the author. Thus, achieving an optimal trade-off between these two conflicting objectives is crucial. In this paper, we propose TAROT: Task-Oriented Authorship Obfuscation Using Policy Optimization, a new unsupervised authorship obfuscation method whose goal is to optimize the privacy-utility trade-off by regenerating the entire text considering its downstream utility. Our approach leverages policy optimization as a fine-tuning paradigm over small language models in order to rewrite texts by preserving author identity and downstream task utility. We show that our approach largely reduce the accuracy of attackers while preserving utility. We make our code and models publicly available.

Problem

• Balancing privacy and utility in authorship obfuscation
• Optimizing trade-off between author anonymity and text quality
• Reducing attacker accuracy while maintaining text utility

Innovation

• Policy optimization for authorship obfuscation
• Balances privacy and text utility
• Rewrites text preserving task utility