privacy through solidarity
play

PRIVACY THROUGH SOLIDARITY Asia J. Biega Rishiraj Saha Roy - PowerPoint PPT Presentation

PRIVACY THROUGH SOLIDARITY Asia J. Biega Rishiraj Saha Roy Gerhard Weikum YOUR PROFILE MEANS PERSONALIZATION SEARCH RECOMMENDATIONS python programming, java, inheritance, nltk Asia J. Biega jbiega@mpii.de YOUR PROFILE


  1. PRIVACY THROUGH SOLIDARITY Asia J. Biega Rishiraj Saha Roy Gerhard Weikum

  2. YOUR PROFILE MEANS … PERSONALIZATION SEARCH RECOMMENDATIONS python … … programming, java, 
 inheritance, nltk Asia J. Biega jbiega@mpii.de

  3. YOUR PROFILE MEANS … YOU SEARCH RECOMMENDATIONS programming Python Programming pottery The 
 ethno jazz Holy Bible mediation The Modern 
 back exercise Libertarian tokyo in august Mein Kampf Asia J. Biega jbiega@mpii.de

  4. YOU CAN’T HAVE YOUR CAKE AND (HAVE YOUR SERVICE PROVIDER) EAT IT TOO … OR Asia J. Biega jbiega@mpii.de

  5. … OR CAN YOU? programming python pottery ethno jazz mediation back exercise tokyo in august Asia J. Biega jbiega@mpii.de

  6. … OR CAN YOU? programming python pottery ethno jazz mediation back exercise tokyo in august Asia J. Biega jbiega@mpii.de

  7. … OR CAN YOU? PRIVACY: programming programming PROFILE FRAGMENTATION pottery pottery ethno jazz ethno jazz mediation mediation back exercise back exercise tokyo in august tokyo in august USER UTILITY: COHERENT CHUNKS Asia J. Biega jbiega@mpii.de

  8. FRAGMENTATION: PRACTICALLY Local user profiles ? Service Provider Asia J. Biega jbiega@mpii.de

  9. MEDIATOR ACCOUNTS PROXY Local user profiles Items Results Mediator Accounts Store 
 no links Item: Result: query 
 ranking 
 rating predicted ratings Service Provider Asia J. Biega jbiega@mpii.de

  10. PRIVACY VS. UTILITY This work PRIVACY USER UTILITY profile scrambled results personalized Asia J. Biega jbiega@mpii.de

  11. PRIVACY VS. UTILITY s u c o f l a u s U PRIVACY USER UTILITY profile scrambled coherent context preserved SERVICE PROVIDER UTILITY minimize analytics loss Asia J. Biega jbiega@mpii.de

  12. PRIVACY VS. UTILITY - STRATEGIES PRIVACY USER UTILITY Random shuffling Results intact Asia J. Biega jbiega@mpii.de

  13. CONTROLLING THE TRADE-OFF COHERENT RANDOM α = 1 α = 0 Entropy Pairwise similarity P ( m | o ) = α · P priv ( m | o ) + (1 − α ) · P util ( m | o ) Profiling-Tradeoff Assignment Asia J. Biega jbiega@mpii.de

  14. MEDIATOR ACCOUNTS: SEARCH QUERY HISTORY programming 
 pottery 
 tokyo in august User U Forwarding programming pottery tokyo in august Similarity: 
 Mediator M topical Language model-based personalization 
 query-doc + doc-mediator Asia J. Biega jbiega@mpii.de

  15. MEDIATOR ACCOUNTS: RECOMMENDERS RATING HISTORY Umiera pi ę kno: 4 
 Koreni: 5 
 Artpop: 1 User U Aggregation Umiera pi ę kno Koreni Artpop Similarity: 
 Mediator M categorical Collaborative filtering Asia J. Biega jbiega@mpii.de

  16. EVALUATION ~900 User profiles Mediator Accounts (Queries synthesized from StackExchange) Profiling-Tradeoff Assignment Asia J. Biega jbiega@mpii.de

  17. EVALUATION ~900 User profiles Mediator Accounts (Queries synthesized from StackExchange) Profiling-Tradeoff Assignment Q: HOW DOES THE PRIVACY-UTILITY TRADE-OFF LOOK LIKE? Asia J. Biega jbiega@mpii.de

  18. EVALUATION: METRICS user u mediator m For each object: ( ( : : Diff* , Ranking over StackExchange answers MODEL EMPIRICAL PRIVACY KL-divergence 
 Entropy 
 topic-level object-level 
 Kendall’s Tau 
 UTILITY (search) 
 Coherence MSE 
 (recommenders) Asia J. Biega jbiega@mpii.de

  19. EVALUATION: OBSERVATIONS TRADE-OFFS (EMPIRICAL) SEARCH SYSTEMS Asia J. Biega jbiega@mpii.de

  20. EVALUATION: OBSERVATIONS TRADE-OFFS (EMPIRICAL) l a e d I SEARCH SYSTEMS Asia J. Biega jbiega@mpii.de

  21. EVALUATION: OBSERVATIONS TRADE-OFFS (EMPIRICAL) f f o - e d a r T l a e d I SEARCH SYSTEMS Asia J. Biega jbiega@mpii.de

  22. EVALUATION: OBSERVATIONS TRADE-OFFS (EMPIRICAL) Utility loss low even for random SEARCH SYSTEMS Asia J. Biega jbiega@mpii.de

  23. EVALUATION: OBSERVATIONS TRADE-OFFS (EMPIRICAL) Alpha influences the variance SEARCH SYSTEMS Asia J. Biega jbiega@mpii.de

  24. EVALUATION: OBSERVATIONS PROFILE DIVERSITY SEARCH SYSTEMS Asia J. Biega jbiega@mpii.de

  25. EVALUATION: OBSERVATIONS PROFILE DIVERSITY Coherent Random SEARCH SYSTEMS Asia J. Biega jbiega@mpii.de

  26. EVALUATION: OBSERVATIONS PROFILE DIVERSITY Coherent high diversity = 
 utility-safe scrambling Random SEARCH SYSTEMS Asia J. Biega jbiega@mpii.de

  27. EVALUATION: OBSERVATIONS Similar observations hold RECOMMENDER SYSTEMS Asia J. Biega jbiega@mpii.de

  28. PRIVACY THROUGH SOLIDARITY THANKS! PROFILING PRIVACY USER UTILITY MEDIATOR ACCOUNTS SEARCH RECOMMENDERS Asia J. Biega jbiega@mpii.de

Recommend


More recommend