What Your Posts Reveal: A Benchmark and Agentic Framework for User-Level Privacy Leakage on Social Media
Researchers have developed SopriBench, a new benchmark designed to evaluate user-level privacy leakage from social media posts. This benchmark utilizes a synthetic dataset derived from real social media accounts and introduces a Privacy Exposure Score (PES) to quantify leakage severity. Additionally, they created Argus, an agentic framework that improves leakage inference by aggregating evidence across multiple posts. AI
IMPACT Introduces novel methods for assessing and mitigating privacy risks in user-generated content.