De#anonymizing,Social,Networks, and,Inferring,Private,Attributes, - PowerPoint PPT Presentation

De#anonymizing,Social,Networks, and,Inferring,Private,Attributes, Using,Knowledge,Graphs, Jianwei Qian Xiang#Yang Li Illinois Tech USTC,/Illinois Tech Chunhong Zhang Linlin Chen BUPT Illinois Tech

Outline Background Prior Work Our Work Conclusion De-anonymizing Social Networks and Inferring 2 Private Attributes Using Knowledge Graphs

Background • Tons of social network data • Released to third-parties for research and business • Though user IDs removed, attackers with prior knowledge can de-anonymize them. → privacy leak De-anonymizing Social Networks and Inferring 3 Private Attributes Using Knowledge Graphs

Attacking Process Prior k.g. De-anonymizing Social Networks and Inferring 4 Private Attributes Using Knowledge Graphs

Prior k.g. Privacy leaked! De-anonymizing Social Networks and Inferring 5 Private Attributes Using Knowledge Graphs

Attack,Stage,1 De#Anonymization Which is Alice? Which is Bob? Direct privacy leak De-anonymizing Social Networks and Inferring 6 Private Attributes Using Knowledge Graphs

Attack,Stage,2 Privacy Inference • Correlations between attributes/users – Higher education => higher salary – Colleagues=> same company – Common hobbies => friends • Infer new info that is not published Indirect privacy leak De-anonymizing Social Networks and Inferring 7 Private Attributes Using Knowledge Graphs

What,Do,We,Want,to,Do? To understand How privacy is leaked to the attacker De-anonymizing Social Networks and Inferring 8 Private Attributes Using Knowledge Graphs

Outline Background Prior Work Our Work Conclusion De-anonymizing Social Networks and Inferring 9 Private Attributes Using Knowledge Graphs

Prior,Work De-anonymize one user Fight Never ending! ◦ Degree attack [SIGMOD’08] ◦ k -degree anonymity ◦ 1-neighborhood attack [INFOCOM’13] ◦ 1-neighborhood anonymity ◦ 1*-neighborhood attack [ICDE’08] ◦ 1*-neighborhood anonymity Assume specific prior knowledge! ◦ Friendship attack [KDD’11] ◦ " # -degree anonymity ◦ Community re-identification ◦ k -structural diversity [SDM’11] De-anonymizing Social Networks and Inferring 10 Private Attributes Using Knowledge Graphs

Prior,Work De-anonymize all the users – Graph mapping based de-anonymization [WWW’07, S&P’09, CCS’12, COSN’13, CCS’14, NDSS’15] Attacker holds an auxiliary SN that overlaps with the published SN Mapping Twitter Flickr De-anonymizing Social Networks and Inferring 11 Private Attributes Using Knowledge Graphs

Limitations • Assume attacker has specific prior knowledge – We assume diverse and probabilistic knowledge • Focus on de-anonymization only. How attacker infers privacy afterwards is barely discussed – We consider it as 2 nd attacking step! De-anonymizing Social Networks and Inferring 12 Private Attributes Using Knowledge Graphs

Outline Background Prior Work Our Work Conclusion De-anonymizing Social Networks and Inferring 13 Private Attributes Using Knowledge Graphs

Goals • To construct a comprehensive and realistic model of the attacker’s knowledge • To use this model to depict how privacy is leaked. De-anonymizing Social Networks and Inferring 14 Private Attributes Using Knowledge Graphs

Challenges • Hard to build such an expressive model, given that the attacker has various prior knowledge • Hard to simulate attacking process, since the attacker has various techniques De-anonymizing Social Networks and Inferring 15 Private Attributes Using Knowledge Graphs

Solution Use knowledge graph to model attacker’s knowledge De-anonymizing Social Networks and Inferring 16 Private Attributes Using Knowledge Graphs

Knowledge Graph • Knowledge => directed edge • Each edge has a confidence score De-anonymizing Social Networks and Inferring 17 Private Attributes Using Knowledge Graphs

What’s Privacy? • Every edge is privacy • Privacy is leaked when $ % e − $ ( (*) > -(*) Say 30% Prior Posterior De-anonymizing Social Networks and Inferring 18 Private Attributes Using Knowledge Graphs

De#Anonymization Mapping A Prior knowledge Anonymized graph 8 % 8 : argmax3456 7 (8 % ,8 : ) 456 7 8 % ,8 : = ∑ 4(5, =) , (>,?)∈7 S is node similarity function De-anonymizing Social Networks and Inferring 19 Private Attributes Using Knowledge Graphs

Node Similarity • Attribute Similarity – Use Jaccard index to compare attribute sets • Relation similarity – Inbound neighborhood – outbound neighborhood – l -hop neighborhood 4 B 5, = = C > 4 > 5,= + C E 4 E 5, = + C F 4 F 5,= 4 5, = = C G 4 G 5, = + 1 − C G 4 B 5, = De-anonymizing Social Networks and Inferring 20 Private Attributes Using Knowledge Graphs

Problem Transformation Mapping => Max weighted bipartite matching Naïve3method: …… 8 % 8 : …… Huge complexity! I % 3I : (millions) De-anonymizing Social Networks and Inferring 21 Private Attributes Using Knowledge Graphs

Top# k Strategy 1 Suppose k=2 Alice 3 8 % 8 : I % ≤ "3I % 3I : (millions) De-anonymizing Social Networks and Inferring 22 Private Attributes Using Knowledge Graphs

How,to,Choose, Top#k Candidates? • Intuition – If two nodes match, their neighbors are also very likely to match. 1 Alice 2 Bob • Perform BFS on 8 % De-anonymizing Social Networks and Inferring 23 Private Attributes Using Knowledge Graphs

Complexity Analysis Time Space Building Bipartite Finding Matching Naïve # # I : I % I : [ I % + I : I % [ I % + I : method Top- k [ " # I % ] [ " # I % # ≪ I % I : strategy Complexity3greatly reduced! De-anonymizing Social Networks and Inferring 24 Private Attributes Using Knowledge Graphs

Tradeoff • " balances accuracy and complexity • " = 10 is enough to achieve high accuracy Accuracy Time 1 50 sr=0.4 sr=0.4 45 sr=0.6 sr=0.6 0.8 sr=0.8 sr=0.8 40 35 0.6 30 0.4 25 20 0.2 15 0 10 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80 k k k k De-anonymizing Social Networks and Inferring 25 Private Attributes Using Knowledge Graphs

Privacy inference Predict new edges in knowledge graph NY Knicks opponent teamInLeague LA Lakers playFor � playFor Nick Young playInLeague teammate Kobe Bryant De-anonymizing Social Networks and Inferring 26 Private Attributes Using Knowledge Graphs

Path,Ranking Algorithm • Proposed by Ni Lao et al. in 2011 for a different topic AIDS Alice • Correlations => “rules” => paths • Logistic regression De-anonymizing Social Networks and Inferring 27 Private Attributes Using Knowledge Graphs

Experiments • Datasets – Google+, Pokec • Steps – Generate 8 : – Generate 8 % – Run the algorithms De-anonymizing Social Networks and Inferring 28 Private Attributes Using Knowledge Graphs

De#Anonymization Results Metrics: accuracy, run time 0.7 40 1 k=5 35 0.6 k=10 0.8 k=15 30 Run Time(s) 0.5 Accuracy Accuracy 25 0.6 0.4 20 0.4 0.3 15 0.2 0.2 10 0.1 Acc 5 De-anonymize about 60% of users 0 Time 0 1 0 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.2 0.4 0.6 0.8 s min w A 80 0.8 70 0.7 Run Time(x10 2 s) 60 0.6 Accuracy 50 40 0.5 RS EG 30 0.4 RW 20 0.3 10 0 0.2 Load Build Match Total Accuracy De-anonymizing Social Networks and Inferring 29 Private Attributes Using Knowledge Graphs

Privacy Inference Results Metrics : hit@k, MRR ( Mean reciprocal rank ) 1.2 6 2.5 12 MRR MRR MRR,RG MRR,RG 1 5 10 2 Hit Hit Hit@10(%) Hit,RG Hit@10(%) Hit,RG 0.8 4 8 MRR(%) MRR(%) 1.5 0.6 3 6 1 0.4 2 4 0.5 0.2 1 Infers much more privacy 2 0 0 0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 sr than random guess sr 5.5 35 30 75 MRR MRR 5 70 30 MRR,RG MRR,RG 4.5 25 65 Hit Hit 4 25 Hit@10(%) Hit,RG 60 Hit,RG Hit@10(%) MRR(%) 3.5 MRR(%) 20 55 20 3 50 2.5 15 15 45 2 40 10 1.5 10 35 1 5 0.5 30 0 0 5 25 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Sample Ratio Sample Ratio De-anonymizing Social Networks and Inferring 30 Private Attributes Using Knowledge Graphs

Outline Background Prior Work Our Work Conclusion De-anonymizing Social Networks and Inferring 31 Private Attributes Using Knowledge Graphs

Conclusion We have • Applied knowledge graphs to model the attacker’s prior knowledge • Studied the attack process: de-anonymization & privacy inference • Designed methods to perform attack • Done simulations and evaluations on two real world social networks De-anonymizing Social Networks and Inferring 32 Private Attributes Using Knowledge Graphs

Future work • Effective construction of the bipartite for large scale social networks • Impact of adversarial knowledge on de- anonymizability • Fine-grained privacy inference on the knowledge graph De-anonymizing Social Networks and Inferring 33 Private Attributes Using Knowledge Graphs

Thank you! Jianwei Qian jqian15@hawk.iit.edu https://sites.google.com/site/jianweiqianshomepage De-anonymizing Social Networks and Inferring 34 Private Attributes Using Knowledge Graphs