presentation slides maya liu

Upload: glenn-micallef

Post on 07-Apr-2018

222 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/3/2019 Presentation Slides Maya Liu

    1/18

    Web Personalization and

    Privacy Concerns

    Maya Liu * COMS E6125 * Spring 2011

  • 8/3/2019 Presentation Slides Maya Liu

    2/18

    What is Web Personalization?

    making web pages more relevant to individual userscookies, data logs, recommendation systems, etc

    source

    http://blog.kissmetrics.com/wp-content/uploads/2010/10/amazon-recommendations.jpg
  • 8/3/2019 Presentation Slides Maya Liu

    3/18

    Why Personalize?

    Saves time and frustration

    source

    http://www.community-learning.org.uk/webs/40/images/Cartoon4.jpg
  • 8/3/2019 Presentation Slides Maya Liu

    4/18

    Why Personalize?

    Saves resources

  • 8/3/2019 Presentation Slides Maya Liu

    5/18

    Why Personalize?

    boosts e-commerce sales

    source

    http://c0162861.cdn.cloudfiles.rackspacecloud.com/20100517_amazon.gif
  • 8/3/2019 Presentation Slides Maya Liu

    6/18

    Current Personalization MethodsWeb-usage mining

    source

    http://maya.cs.depaul.edu/~mobasher/personalization/architecture.gifhttp://maya.cs.depaul.edu/~mobasher/personalization/architecture.gif
  • 8/3/2019 Presentation Slides Maya Liu

    7/18

    Current Recommendation Systems

    Content-Basedusers and web sites are

    modeled as objects with

    attributes

    think database entries

    recs based on user historyand attributesnot useful when there arelarge amounts of

    unstructured data

    Case-Basedan improvement overcontent-based systemused in e-commerceeliminates unstructureddatarequires same fields forobjects of the same typenot good for objects with

    unstructured datapotentially lack diversityin results

  • 8/3/2019 Presentation Slides Maya Liu

    8/18

    Privacy Concerns

    source

    http://www.personal.psu.edu/gtk5002/privacy1.gifhttp://www.personal.psu.edu/gtk5002/privacy1.gif
  • 8/3/2019 Presentation Slides Maya Liu

    9/18

    Privacy Survey, c.2000 source: [8]

  • 8/3/2019 Presentation Slides Maya Liu

    10/18

    Challenges to Preserving Privacypersonal definitions of privacy differ

    regional, cultural, and background differencesregional laws and regulations

    source

    http://www.toothpastefordinner.com/081208/online-privacy-advocate.gif
  • 8/3/2019 Presentation Slides Maya Liu

    11/18

    Current Privacy Solutions

    Largest permissible dominatorRegion-specific versionsPseudonymous personalization

  • 8/3/2019 Presentation Slides Maya Liu

    12/18

    Current and Future Improvements

    Pseudonymous Personalization

    must satisfy the following criteria:unidentifiablelinkable for the personalized system but not thirdpartiesunobservable for third parties

    pseudonymous users AND servers

  • 8/3/2019 Presentation Slides Maya Liu

    13/18

    Current and Future Improvements

    Client-side Personalization

    store personal data on client computers instead of serversserver leaks pose less risks and potentially increases userwillingness to provide personal datachallenges

    aggregate data calculationsrecommendation software must be sent to clients,providing opportunities for reverse-engineering

  • 8/3/2019 Presentation Slides Maya Liu

    14/18

    Case Study: SUGGEST

    "Novel approach to implementing Web personalization as a

    single online module"uses "usage clusters" to represent sessionsranks page importance by order of visit from starting pagerelies on aggregate data to weed out outlier dataconstructs directed graph from data points and use forrecommendations

    2007, [3]

  • 8/3/2019 Presentation Slides Maya Liu

    15/18

    Case Study: SUGGESTPros:

    stores data in adjacency matrix, and can change itssize and recommendation level based on constraintsno personal data collected, and uses link informationalready provided on many websites

    Cons:

    restricted to page recommendationsexploitable, similar to Google bombing

    2007, [3]

  • 8/3/2019 Presentation Slides Maya Liu

    16/18

    Future Research Directionsuser anonymitypassive data-gatheringinferring preferences from aggregate data, not specificuser's history

    source

    http://scienceblogs.com/omnibrain/upload/2006/12/anonymous.jpg
  • 8/3/2019 Presentation Slides Maya Liu

    17/18

    Questions? Comments?

    source

    http://farm5.static.flickr.com/4072/4678884792_acfeba9fe5.jpg
  • 8/3/2019 Presentation Slides Maya Liu

    18/18

    Reference[1] Alfred Kobsa, Privacy-enhanced web personalization, The adaptive web: methods andstrategies of web personalization, Springer-Verlag, Berlin, Heidelberg, 2007[2] Michael J. Pazzani , Daniel Billsus, Content-based recommendation systems, The adaptive web:methods and strategies of web personalization, Springer-Verlag, Berlin, Heidelberg, 2007

    [3] Ranieri Baraglia , Fabrizio Silvestri, Dynamic personalization of web sites without userintervention, Communications of the ACM, v.50 n.2, p.63-67, February 2007

    [4] Barry Smyth, Case-based recommendation, The adaptive web: methods and strategies of web

    personalization, Springer-Verlag, Berlin, Heidelberg, 2007

    [5] Magdalini Eirinaki , Michalis Vazirgiannis, Web mining for web personalization, ACMTransactions on Internet Technology (TOIT), v.3 n.1, p.1-27, February 2003

    [6] Bamshad Mobasher , Robert Cooley , Jaideep Srivastava, Automatic personalization based onWeb usage mining, Communications of the ACM, v.43 n.8, p.142-151, Aug. 2000

    [7] Teltzrow, M. and Kobsa, A.: Impacts of User Privacy Preferences on Personalized Systems: aComparative Study. In: Designing Personalized User Experiences for eCommerce, Karat, C.-M.,Blom, J., and Karat, J., Eds. Dordrecht, Netherlands: Kluwer Academic Publishers (2004) 315-332,DOI 10.1007/1-4020-2148-8_17.