Keep It Private: Unsupervised Privatization of Online Text
This work addresses privacy protection for online text users, but it is incremental as it builds on existing authorship obfuscation techniques with a more advanced method.
The paper tackles the problem of protecting user privacy in online communications by automatically rewriting text to hide authorship, introducing a framework that fine-tunes a large language model via reinforcement learning to balance soundness, sense, and privacy. The result shows that the method maintains high text quality and successfully evades several automated authorship attacks on a large-scale test set of English Reddit posts by 68k authors.
Authorship obfuscation techniques hold the promise of helping people protect their privacy in online communications by automatically rewriting text to hide the identity of the original author. However, obfuscation has been evaluated in narrow settings in the NLP literature and has primarily been addressed with superficial edit operations that can lead to unnatural outputs. In this work, we introduce an automatic text privatization framework that fine-tunes a large language model via reinforcement learning to produce rewrites that balance soundness, sense, and privacy. We evaluate it extensively on a large-scale test set of English Reddit posts by 68k authors composed of short-medium length texts. We study how the performance changes among evaluative conditions including authorial profile length and authorship detection strategy. Our method maintains high text quality according to both automated metrics and human evaluation, and successfully evades several automated authorship attacks.