Audience-conditioned privatization for private text generation
Original title: Who's Asking? Audience-Conditioned Privatization
Ext-Classify-Priv framework classifies attributes by audience to preserve only contextually sensitive information in differentially private synthetic data releases.
Ext-Classify-Priv extracts information types from corpora and classifies private attributes per audience using contextual integrity. It protects only those with anonymization or feature-DP synthetic generation. On Reddit data with three audiences, audience-specific releases are more faithful than agnostic ones. It improves MAUVE scores on bioRxiv and real estate datasets over baselines.
- Audience-specific releases more faithful than single release
- Classifies private attributes per audience
- Uses anonymization or SDG with Feature-DP
- Improves MAUVE from 0.097 to 0.420 on bioRxiv data
- Improves MAUVE from 0.698 to 0.959 on real estate data
From the original article
1 min read at arxiv.orgThe abundance of user-generated conversational data makes the question of how to responsibly derive insights from confidential data without putting the individuals in it at risk ever more important. However, the majority of existing differentially private (DP) synthetic text generation (SDG) and anonymization approaches ignore the context of the data's downstream usage, which results in loss of important information.
By Daniil Filienko · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.