Personalized hypermedia presentation techniques for improving online customer relationships

1
Personalized hypermedia presentation techniques
for improving online customer relationships

• Kobsa, Koenemann, and Pohl

Presenters Stacy Tang and Matt Yeh
2
Outline

• Introduction
• Input data
• Acquisition methods
• Representation and secondary inferences
• Adaptation production
• Conclusion

3
Outline

• Introduction
• Input data
• Acquisition methods
• Representation and secondary inferences
• Adaptation production
• Conclusion

4
IntroductionWhy personalization?

• providing value to customer
• Brick and Mortar
• personal service
• tailored products

5
IntroductionWhy use the web for personalization?

• Collect large amount of data
• Rapid updates
• World-wide and 24/7
• Dynamic creation of content

6
IntroductionWhy personalize on the web?

• page views
• length of page views
• new customers
• visitors
• revenue

7
IntroductionHow the Internet fits in
Sales Cycle
Establish and strengthen brand
Online ordering and purchasing
During
Pre
Post
Reassure customer and product support
8
Introduction Definition

• Personalized Hypermedia Application
• An interactive system that allows users to
  navigate a network of linked hypermedia objects
  (i.e. web pages) and adapts the content structure
  and/or presentation of the networked hypermedia
  objects to each individual users
  characteristics, usage behavior and/or usage
  environment.

9
Outline

• Introduction
• Input data
• Acquisition methods
• Representation and secondary inferences
• Adaptation production
• Conclusion

10
Input DataUser data

• Information about personal characteristics of the
  user
• Demographic
• Knowledge
• Skills and capabilities
• Interests and preference
• Goals and plans

11
Input DataUser data - demographics

• Objective facts
• record
• geographic
• characteristics
• lifestyle
• registration

12
Input DataUser data - user knowledge

• knowing what
• Adjust the presentation based on user knowledge
• expert not bored by unnecessary details
• novice not confused by details they dont
  understand

13
Input DataUser data - user knowledge ex
14
Input DataUser data - skills capabilities

• skills - knowing how actions that the user is
  familiar with
• capabilities - actions that user is able to
  perform

15
Input DataUser data - interests and preferences

• Align content with user interests
• Important in recommendation systems

16
Input DataUser data - goals and plans

• Plan-recognition
• Facilitate interaction

17
Input DataUsage data

• Directly observed
• ways users interact with a system
• can directly lead to adaptation
• General regularities
• further process the above to deduce information
  about the user

18
Input DataObservable usage - selective actions

• Clicking on a link as an indicator for
• interest ( only)
• unfamiliarity ( only)
• preference

19
Input DataObservable usage - other interactions
usage and indicator for user interest
Viewing time of page
Save document, print document, bookmarking
page, forward story by email
Explicit ratings (i.e., Amazon)
putting items in shopping cart
20
Input DataUsage data - finding regularities

• Process usage data to find
• Frequency
• Situation-based correlations
• Action sequences

21
Input DataEnvironment data

• Software
• browser, platform
• plug-ins
• Java and Javascript
• Hardware
• bandwidth
• processing speed
• display
• input
• Locale
• location
• characteristics of location

22
Outline

• Introduction
• Input data
• Acquisition methods
• Representation and secondary inferences
• Adaptation production
• Conclusion

23
Acquisition Methods User Acquisition Methods

• Methods to obtain data that can be input into
  personalized hypermedia
• User information -gt user model
• Usage Information -gt usage model
• Environment

24
Acquisition MethodsUser Model Acquisition
Methods

• Strategies for obtaining data about user
  characteristics
• Active methods
• Passive Methods

25
Acquisition Methods User Supplied Information

• Obvious strategy is to have user supply info
• Initial Interview
• Registration Process
• Examples
• Soccernet.com
• My.Yahoo.com

26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
Acquisition MethodsProblems with Interviews

• Self-assessment may be error-prone
• Solution Indirect assessment

30
Acquisition Methods Indirect Assessment

• Website where we want to acquire a user
  characteristic
• Users expertise in speaking English
• We can ask user
• Better method may be to determine expertise
  indirectly

31
Acquisition Methods Problems with Interviews,
cont.

• Paradox of the active user
• User anxious to begin immediate task and are too
  busy for setup
• Doing setup may save user time later

32
Acquisition Methods Problems with Interviews,
cont.

• Solutions to this problem
• Let the user initiate setup
• Fold setup into interaction gradually
• Automate setup

33
Acquisition Methods Passive Acquisition

• Acquisition where interaction is not initiated
  with user
• Less disturbing or annoying
• Passive Acquisition Methods
• Acquisition rules
• Plan Recognition
• Stereotype Reasoning

34
Acquisition Methods Acquisition Rules

• Heuristics or Inference rules
• Generate assumptions about user given available
  information
• Example If user wants to know concept X, we
  assume that user does not know concept X

35
Acquisition Methods Acquisition Rules

• Example We want to know the users level of
  experience with a program
• We can accomplish this with inference rules based
  on knowledge of when the user last used the
  program
• If the user has been away too long
• Downgrade experience level by 1.
• If the user has used the system long enough since
  the last update
• Upgrade experience level by 1.

36
Acquisition MethodsPlan Recognition

• Reasoning about user goals action sequences
  user performs to achieve them
• Monitor user action to ID user plan/goal
• Modify our program to help user efficiently
  achieve those goals

37
Acquisition MethodsPlan Recognition

• Microsoft XP monitors the applications the user
  most frequently uses

38
Acquisition Methods Stereotype Reasoning

• Stereotype reasoning for hypermedia is a method
  that works like everyday stereotyping

39
Acquisition Methods Stereotype Reasoning

• We create categories of users and maintain a body
  of info true for users in each category
• We have triggers for assigning users to
  categories
• Then we can make assumptions about user based on
  category membership
• Example Searching for info about childcare
  activates a parent stereotype that we use to make
  predictions about user characteristics

40
Acquisition Methods Usage Acquisition Methods

• Acquiring usage info seems to be an easier task
• We observe and record what the user does
• Simply observing user behavior may not be enough

41
Acquisition Methods Usage Acquisition Methods

• Often we want to know the context in which a user
  performs particular actions
• We can then use machine learning strategies to
  predict user actions given a certain scenario
• Situation/Action learning

42
Acquisition MethodsEnvironmental Data

• We may want to know about context in which
• user interacts with a system
• Software Information
• Browser type
• Affects how hypermedia appears
• Determine browser type through header of http
  request
• Special programs to determine browsers type

43
Acquisition Methods Environmental Data
44
Acquisition Methods Environmental Data

• Bandwidth
• Difficult to detect
• Special software to predict download times
• Prediction can be used to adapt page composition
• Hardware
• Difficult to get
• Can sometimes assume from browser

45
Acquisition Methods Environmental Data

• User location
• Often we want to tailor hypermedia based on user
  location
• Consider navigation system
• For such mobile devices
• Electromagnetic fields (GPS, Bluetooth, radio,
  etc.)
• Ultrasound
• IR and optical recognition
• For stationary networked devices location is
  often stored in a database

46
Outline

• Introduction
• Input data
• Acquisition methods
• Representation and secondary inferences
• Adaptation production
• Conclusion

47
Representation and Secondary Inference

• Store user information in a way that is useful
• Can use simple methods
• Example Maintain a list of feature-value pairs
  like
• CONCEPT X KNOWN / CONCEPT X NOT KNOWN

48
Representation and Secondary Inference

• Some systems have higher demands
• Need to represent information to make inferences
  based on initial acquisition results
• Secondary Inferences

49
Representation Secondary InferenceDeductive
Reasoning Strategies

• Use a system based on logic to represent info and
  make inferences
• Logic-based formalisms
• Propositional logic
• Modal logic

50
Representation Secondary Inference
Logic-based approach Concept Hierarchy
thing
fish
mammal
shark
whale
orca
humpback
51
Representation Secondary Inference
Logic-Based approaches

• Shortcomings
• Method is rigid. Does not deal with changes to
  user model well
• Does not deal with uncertainty well
• May not be sure contents of user/usage model are
  accurate
• For example, we might be 60 sure that the user
  knows that a shark is a fish

52
Representation Secondary Inference Inductive
Reasoning Learning

• In previous examples, we wanted to draw specific
  assumptions about users
• Use specific observations to draw general
  conclusions
• In domain of customer relation management we are
  most often concerned with creating a general
  interest profile

53
Representation Secondary Inference Interest
Profile

• Representation of users general preference or
  affinity for object based on features of object
• Example Movies

Action Movies
54
Representation Secondary Inference
Techniques to Acquire an Interest Profile

• Machine Learning techniques
• Neural Networks
• Example Neural net to assemble a interest
  profile about websites
• Create network based on features of website
• Train network with the a users ratings of
  websites
• Network "stores" the interest profile
• Network will predict if a new website will be
  interesting to the user

55
Representation Secondary Inference Problems
with feature-driven inductive approaches

• Not easy to parse out features of some objects
  (e.g. multimedia objects)
• Training period (as in neural net example) may
  not be possible.
• Interest in object may not depend on features

56
Representation Secondary Inference
Analogical Reasoning

• Reasoning based on similarity of users
• One technique is Clique-based filtering

57
Representation Secondary Inference
Clique-based Filtering

• Adapt to the individual user based on the
  behavior of similar users
• Interest neighbors
• The set of similar users constitute an implicit
  profile of user
• Make predictions based on implicit profile

58
Representation Secondary Inference
Clique-based filtering

• Example Amazon.com looks for users who have made
  similar purchases and makes predictions about
  other products you may like

59
Outline

• Introduction
• Input data
• Acquisition methods
• Representation and secondary inferences
• Adaptation production
• Conclusion

60
Adaptation ProductionAdaptation of content

• Functions of adapting content
• optional explanation
• optional detailed information
• personalized recommendations
• optional opportunistic hints

61
Adaptation ProductionAdaptation of content

• Techniques of adapting content
• page variants
• fragment variants
• fragment coloring
• adaptive stretchtext
• adaptive natural-language generation

62
Adaptation ProductionAdaptation of content
example

• CNN.com page variant

63
Adaptation ProductionAdaptation of content
example

• MyYahoo fragment variant

64
Adaptation ProductionAdapt presentation and
modality

• Change of format and layout
• Change of modality

65
Adaptation ProductionAdapt presentation example

• MyYahoo personalize format

66
Adaptation ProductionAdapt presentation example

• MyYahoo personalize layout

67
Adaptation ProductionAdapt modality example

• Map
• directions
• Text only
• directions

68
Adaptation ProductionAdaptation of structure

• Functions of structure adaptation
• recommendations
• products, information, and navigation
• orientation and guidance
• personalized overview maps, guided site tours
• personal views and spaces
• bookmarks

69
Adaptation ProductionAdaptation of structure

• Techniques for structure adaptation
• link sorting
• link annotation
• link hiding and unhiding
• link disabling and enabling
• link removal/addition

70
Adaptation ProductionAdaptation of structure
example
71
Adaptation ProductionAdaptation of structure
example
Recommend products with link additions
72
Outline

• Introduction
• Input data
• Acquisition methods
• Representation and secondary inferences
• Adaptation production
• Conclusion

73
Conclusions ProspectsPersonalization
applications

• Where will personalization be used?
• Public websites
• keeping visitors
• turning visitors into customers
• making visitors return
• Personalization not always needed, and will not
  make human sales people obsolete
• Websites where customers can ask for human
  assistance can be effective

74
Conclusions Prospects Personalization
Applications

• Nordstrom.com includes a link showing user how to
  contact a sales expert

75
Conclusions ProspectsPersonalization
Applications

• "Walk-up and use" kiosks found in fairs,
  exhibitions, showrooms
• Mobile Devices
• Phones
• PDAs
• Car-mounted Devices
• Universal Access systems
• Hypermedia personalized to meet needs of special
  users
• E.g., those with disabilities

76
Conclusions Prospects Recommendations for
Personalization

• Remember the "Paradox of the active user
• Avoid lengthy registration process
• Expose user to content immediately
• Offer adaptation as an option
• Allow user to correct or undo adaptations

77
Conclusions Prospects Recommendations, cont.

• Log user navigation at page level
• critical in site design
• Logging and personal info acquisition in general
  leads to privacy concerns
• must be addressed proactively
• tell user what is being done with personal info
• tell them how providing personal info improves
  user experience
• if possible, allow user to opt out of logging