Preserving the Privacy of Sensitive Relationships in Graph Data - - PowerPoint PPT Presentation

Data Anonymization

Preserving the Privacy of Sensitive Relationships in Graph Data

Privacy Analyst Valuable Data! Data publisher Public No privacy breaches!

Data representation? Privacy breach? Value of data?

Motivation

Work by: Elena Zheleva, Lise Getoor, U of Maryland

Data representation

• Most approaches to data anonymization

assume that the original data records are in a single flat table

• Identity disclosure
• Sensitive attribute disclosure

vs.

20740 M 24 John 20760 M 23 Ian 83201 F 29 Halle 20640 F 24 Gia 83222 M 31 Fabio 83230 F 28 Emma 83209 M 29 Don 20742 M 24 Chris 83201 M 25 Bob 20740 F 21 Ana

Preserving the Privacy of Sensitive Relationships in Graph Data

• Often, data is more complex, and it

exhibits various relationships

• E.g., friends, classmates, etc.

Work by: Elena Zheleva, Lise Getoor, U of Maryland

Possible privacy breaches

• Sensitive relationship disclosure
• Sensitive relationship: one that has to

remain private given the anonymized data

• Goal: anonymize data (nodes and links) in a way to hide sensitive relationships
• We propose 5 anonymization strategies (see our poster for details)
• Compare the sensitive relationship disclosure in the anonymized data
• Privacy breaches depend on the adversary domain knowledge
• Assume the adversary can compute the probability of sensitive links
• Probability is based on observed properties of the data

Preserving the Privacy of Sensitive Relationships in Graph Data

Anonymizing graph data

Work by: Elena Zheleva, Lise Getoor, U of Maryland

Link re-identification “Sensitive” link examples

Disease data

father-of has hypertension

Social network data

friends

Search data

Query 2:

“myrtle beach golf course job listings”

Query 1:

“how to tell if your wife is cheating on you”

same-user