Keep It Private: Unsupervised Privatization of Online Text

arXiv:2405.10260v1 Announce Type: new
Abstract: 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 …

abstract arxiv author authorship communications cs.ai cs.cl edit hide however identity literature narrow nlp people privacy protect text type unsupervised