User Modeling

1
User Modeling Adaptive User Interfaces
Dialogue of System and Theorygrace de la flor
2
Introduction

• What is a user model
• Computer perspective
• What is an adaptive user interface
• Human perspective
• Review of Web sites
• Framework for evaluation

3
Brief History

• Early 1990s
• Research based
• Relative isolation
• 1994
• First workshop on adaptive hypermedia
• 1996
• Turning point - new research teams PhDs
• World Wide Web
• 1997
• User Modeling and User-Adapted Interaction
  Journal
• Today
• Handheld guides
• Location based mobile content
• Web sites

4
Why have adaptive systems?

• To enhance the user experience
• Information filtering
• Individual interests
• Shared interests
• Tailored content navigation

5
Synonyms for User Modeling

• Personalisation
• Filtering systems
• Adaptive systems
• Adaptive user interfaces
• Recommender systems

6
Personalisation Techniques

• Manual
• Designer based changes from feedback and log
  analysis
• Content filtering
• What the user is interested in
• subject specific, location specific
• Collaborative filtering
• Who else is interested
• Recommender systems, Most read articles, most
  viewed photos

7
User Modeling Adaptation
Brusilovsky P., Maybury M. From adaptive
hypermedia to the adaptive web. Pages 30 33.
Communications of the ACM May 2002/Vol. 45, No. 5
8
System Knowledge Acquisition

• Implicit data
• Data mining
• Navigation, content
• Explicit data
• User input
• User preferences, profile

9
The computer perspective

• Finding patterns in pre-existing data
• Using Implicit Explicit information
• Collect data
• Process data
• Apply user model
• Present the adaptive interface

10
Implicit Data

• Server logs Cookies
• Patterns of use
• IP address
• Time stamp
• Referring URL
• Operating system
• User ID session number
• Number of visit/pages viewed
• Data collection
• Pattern discovery
• Automated user model

11
Explicit Data

• User Profiles Preferences
• Supplied data
• Age
• Sex
• Location
• Content preferences
• Layout preferences
• Shopping history
• User input
• Pattern discovery
• Informed user model

12
Pattern Discovery

• Machine Learning Statistical Models
• Clustering
• Grouping users with common browsing behaviour
• Grouping web pages with similar content
• Classification
• Model the behaviour of users and the
  classification of Web pages
• Keyword or page views represent user interests
• Pages that share links have a relationship
• Association discovery
• Represent causal relations about a number of
  variables
• Bayesian networks - Statistical inference in
  which probabilities are discovered not in terms
  of frequencies but, as degrees of belief
• Gaining popularity in the Artifical Intelligence
  community
• Sequential pattern discovery
• The element of time, event sequences
• Markov model - the distant past is irrelevant
  given knowledge of the recent past
• Navigation patterns to predict future visits
  based upon past page views

13
Recommender Models

• Action-to-Item affinities
• Based upon past queries
• page tracking, wish lists
• Item-to-Item affinities
• Between products user expressed interest in
• you like item x you may like item y
• User-to-User affinities
• Between a user and like-minded users
• users who liked x, also liked y
• --------------------------------------------
  --------------------------------------------------
  --------------------------------------------------
  ------------------------------------------
• Fink J, Kobsa A, A Review and Analysis of
  Commercial User Modeling Servers for
  Personalization on the World Wide Web, UMUAI 10,
  pp. 209-249, (2000)

14
Stereotypes
Assumptions made about users based upon
navigation history, shopping patterns, profile
settings

• Information about users
• User attributes
• Information about things
• Object definition
• Object attributes

15
Objects Attributes

• User and Domain
• User Attribute Domain Object
• Users who have x attribute will dis/like y
  objects
• Domain Object User Attribute
• Identifying experts associate domain objects
  with users
• User Attribute User Attribute
• Recommender systems users sharing same
  attributes, share similar interests
• Domain Object Domain Object
• Object to object association liked movie x,
  youll like movie y
• Domain Attribute Domain Object
• Attribute to object Dont like violence in
  films, wont like x film
• ------------------------------------------------
  --------------------------------------------------
  --------------------------------------------------
  ------------------------------------------
• Kay Judy, Lies, damned lies and stereotypes
  pragmatic approximations of users

16
Summary of User Modeling

• Content filtering
• user profiles, navigation shopping history
• Collaborative filtering
• associations between similar users
• Implicit and explicit data
• data tracking, purchase history user profiles,
  preferences
• Pattern discovery
• algorithms used for machine learning
• Stereotypes
• assumptions about users

17
The human perspective

• Presenting user models in adaptive user
  interfaces
• Web site evaluation criteria
• User salutation
• Saving (searches,items,bookmarking)
• Guidance (recommendations)
• Customisation (layout, content, results sorting)
• Task support (automated searches, bid
  negotiator,automatic email or downloading)
• --------------------------------------------------
  --------------------------------------------------
  --------------------------------------------------
  ------------------------------------------
• Pierrako D, et al, Web Usage Mining as a Tool
  for Personalization A Survey, UMUAI 13, pp.
  311-372, (2003).

18
Adaptive vs. Adaptable

• Design time vs. Use time
• Adaptive
• By the system itself to task user definition
• Knowledge contained in the system knowledge
• Little or no effort from user
• Loss of user control -
• Adaptable
• User can change functionality definition
• Knowledge extended by user knowledge
• User in control
• User must do substantial work -
• -----------------------------------------------
  --------------------------------------------------
  --------------------------------------------------
  -------------------------------------------
• Fisher G, User Modeling in Human-Computer
  Interaction, UMUAI 11, pp. 65-86, (2001).

19
(No Transcript)
20
(No Transcript)
21
How its done

• Login, Choose content, Choose layout, Choose
  colour, Add/delete pages
• -http//help.yahoo.com/help/uk/my/my-05.html
• Content filtering
• Subject specific, location based
• Explicit data, user preferences
• Issues
• No access to Yahoo Groups
• No access to My Profile public web page

22
Privacy Policy

• Information Collection and Use
• Once you register with Yahoo! and sign in to our
  services, you are not anonymous to us
• Explicit Data
• name, email address, birth date, gender, post
  code, occupation, industry, and personal
  interests.
• Implicit Data
• IP address
• Cookies
• Use
• customise advertising and content you see, based
  on registration and activity at Yahoo!
• Third party exchange

23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
How its done

• Based on the items you've purchased from and the
  behaviour of other customers who've bought the
  same items, Recommendations change immediately
  when you purchase or rate a title, Can change
  recommendation by clicking not interested or
  own it -http//www.amazon.co.uk
• Collaborative filtering
• Association by user and object
• Explicit data, shopping history
• Implicit data, stereotypes
• Issues
• Cannot disable recommender
• Creates inaccurate stereotypes

28
Privacy Policy

• Controllers of Personal Information
• Any personal information provided or to be
  gathered by Amazon.co.uk is controlled primarily
  by Amazon.com
• Explicit Data
• Your account, About you, Reviews, Wish lists,
  Refer-a-friend
• Implicit Data
• IP address, Web browser, Operating system
• Use
• Optimise services
• Third party exchange

29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
How its done

• new algorithms that dynamically reorder results
  by weighting the interests you enter in your
  profile. When you move the slider, it
  recalculates and rearranges the results to add
  more or less emphasis on your profile
  information. -http//labs.google.com/personalized/
  faq.html
• Content filtering
• Subject specific
• Explicit data, user preferences
• Issues
• Weighting at times not effective or relevant
• Doesnt allow for multiple saved profiles
• Cannot save searches
• Cant email results
• Cant set up alerts

33
Privacy Policy

• Information Collection and Use
• Google will not disclose its cookies to third
  parties except as required by a valid legal
  process such as a search warrant, subpoena,
  statute court oder
• Explicit Data
• query
• Implicit Data (with each query)
• time of day, browser type, browser language, and
  IP address
• Use
• to verify our records and to provide more
  relevant services to users
• advertisements
• Third party exchange

34
Privacy and User Modeling

• Country of origin
• Transborder flow of personal data
• Use of data
• Purpose, storage time, third party sharing
• Policies
• Data oriented - what data, time
• Method oriented - automated models
• User inspection of data
• Inform users
• Opt in/out mechanism
• Provide non-personalised version
• Provide anonymous version
• Provide security
• Tailor user model to privacy preferences
  legislation

35
Privacy Questions to Ask

• Disable and delete cookies
• Will the system still work?
• Sharing purchase history or wish lists
• Is it optional?
• Using my data to create collaborative filters
  without my consent
• Can I opt out?
• Third party use
• Who else is accessing my data?
• Privacy and Use
• Are rights negated upon use?

36
References

• Brusilovsky, P. Adaptive Hypermedia, UMUAI 11,
  pp. 87-110, (2001).
• Fink J, Kobsa A, A Review and Analysis of
  Commercial User Modeling Servers for
  Personalization on the World Wide Web, UMUAI 10,
  pp. 209-249, (2000).
• Fisher G, User Modeling in Human-Computer
  Interaction, UMUAI 11, pp. 65-86, (2001).
• Kay Judy, Lies, damned lies and stereotypes
  pragmatic approximations of users
• Kobsa A. Tailoring Privacy to User Needs.
  Keynote, 8th International Conference on User
  Modeling (2001).
• Pierrako D, et al, Web Usage Mining as a Tool
  for Personalization A Survey, UMUAI 13, pp.
  311-372, (2003).
• Zukerman I, et al, Predictive Statistical Models
  for User Modeling, UMUAI 11, pp. 5-18, (2001).