GraphProfiler reveals specific posts behind private data guesses
GraphProfiler: Source-Linked Sensitive Attribute Inference via Personal Knowledge Graphs
Computation and LanguageCryptography and Security
Summary
Sensitive personal details like age or income can be guessed by looking at a user's many online posts combined. The authors show that large language models can do this well but don't explain which posts lead to these guesses. They created GraphProfiler, which links each guess to specific posts by building a personal knowledge graph. This helps people see exactly which posts reveal their private info, allowing more focused privacy protection.
What this means in practice
- •For privacy tool developers: Build tools that automatically identify the exact user posts leaking private attributes for targeted content redaction.
- •For social media platform operators: Integrate profiling tools that alert users and moderators about posts strongly contributing to sensitive data inference risks.
Authors
Ahmed Sohair Khan, Estrid He, Chenglong Ma, Monica Wachowicz, Elham Naghizade
Abstract
Sensitive attributes such as age, income, and occupation can be inferred from user-generated content by aggregating indirect cues across many ordinary posts. LLM-based profilers can perform this aggregation automatically and with high accuracy, which makes large-scale personal attribute inference a major privacy threat. Existing LLM-based profilers, however, offer limited insight into which specific posts, concepts, and relationships made an inference possible, which is key to targeted privacy mitigation, i.e., redacting or rewriting only the few posts that actually leak an attribute, rather than perturbing entire histories. We introduce GraphProfiler, an auditable LLM-based profiler that represents each user's post history as a source-linked personal knowledge graph where nodes and edges trace back to the originating post and resolves attribute predictions to cited graph records and source texts. GraphProfiler reaches 86.7% attack success rate on the eight-attribute SynthPAI benchmark, within two points of strong text-only baselines, and 84.6% on PANDORA, while citing supporting evidence for over 98% of predictions. Our controlled ablation experiments provide evidence that the cited posts contribute to attack success, as removing them reduces the attack success rate substantially more than removing an equal number of random posts.