User modeling 2

1
User modeling 2
2
System asks questions
User answers
Stereotypes are determined
User model is constructed
System observes user actions
System output adapted to user model
User uses system
3
What is being modelled?

• Knowledge
• Interests
• Goals and tasks
• Background
• Individual traits
• Affective state Separate lecture
• Context of work

4
Knowledge

• Important in Adaptive Educational Systems
• Also used in other Adaptive Hypermedia
• Changeable knowledge acquired and forgotten
  (both in and outside system use)
• Challenge deal with changing knowledge

5
Knowledge Scalar Models

• Estimate knowledge on a scale
• Quantitative 0-10
• Qualitative good, average, poor, none
• Often by self-reporting (ask the user)
• Can result in stereotype (e.g. novice, expert)
• Adaptations based on stereotype

6
Knowledge Structured Models

• Scalar models have low precision, and knowledge
  in large area can vary
• Structured models represent knowledge is
  fragments of the domain
• Overlay model users knowledge as a subset of
  experts knowledge

7
Knowledge Example of Overlay Model
South America
Country
Country
Argentina
Brazil
Capital
Language
Language
Capital
Portuguese
Spanish
Buenos Aires
Brazilia
8
Knowledge Overlay Model

• In the example, the user had knowledge or no
  knowledge (binary)
• Can use more fine-grained
• 80 confident that user has knowledge
• user knows 75 of topic
• users knowledge is good
• Often knowledge level certainty level

9
Knowledge Bug Model

• Extension of overlay model
• Also models misconceptions
• Problem if the user makes a mistake, which
  misconceptions are causing it?

10
Interests

• Important in Information Retrieval and
  Recommender Systems (future lectures)
• Interests may change over time, but some
  interests may be quite stable

11
Interests Keywords Vector

• Popular approach weighted vector of
  keywords (programming, 6), (java, 5), (php, 2),
  ..
• Weights can be provided by user, or learned over
  time, or initiated from stereotype

12
Interests Concept Level Model

• Overlay model for interests
• Different areas of interest are modelled
  separately
• Semantic links can compensate for scarcity, e.g.
  Aberdeen is in Scotland, so if not
  interested in Scotland then probably not
  interested in Aberdeen

13
Interests Example of Concept Level Model
Sports news
8
4
Football
Rugby
Cricket
Motorsport
F1
BSB
Premier League
SPL
World Cup
9
14
Interests Concept Level Model

• To adapt, need to know how items (for example,
  news stories) map onto concepts
• Option 1 Manually index
• Option 2 Automatically index
• Option 3 Hybrid ( combination)

15
Goals or Tasks

• Why the user wants to use the system
• For example
• Information need (e.g. in search engine)
• Learning goal (e.g. in educational system)
• Goal of work (e.g. in text editor)
• Very changeable

16
Goal modelling

• Current goal often modelled in similar way to
  overlay models
• System recognises predefined goals / tasks
• Can be hierarchically structured

Write a letter
Insert address
Add signature
Write body
Add greeting
17
Goal modelling

• Often system assumes that user has one goal at a
  time
• Tries to recognise goal and adapt
• For example
• Re-order result list in search engine
• Make links bigger that may interest user
• Format page for a lettter

18
Background

• Previous experience, outside the system
• For example job, (prior) education, experience
  in related areas, language ability
• Very stable
• Hard to determine automatically, so ask
• Only model what you can use...
• Often used for content adaptation (e.g. Medical
  information different for nurses, doctors,
  patients)

19
Individual Traits

• Features that define user as an individual
• Stable
• Often determined using psychological tests
• Possibilities
• Personality traits
• Cognitive styles
• Cognitive factors
• Learning styles

20
Individual Traits Personality

• Five Factor Model (Big Five)
• Conscientiousness being careful, organised
• Openness to experience being curious, liking to
  try new things
• Extraversion being assertive, outgoing, chatty
• Agreeableness being pleasant, friendly, helpful
• Neuroticism enduring tendency to experience
  negative emotional states (e.g. anxiety, anger,
  guilt)

21
Individual Traits Personality

• Self-efficacy belief in abilities to reach a
  goal
• Locus of control belief about the extent to
  which behaviours influence success
• Goal-orientation, e.g. oriented towards learning
  or towards looking good compared to others
• Self-esteem overall evaluation of self-worth
• ETC

22
Individual Traits Cognitive Styles

• Individually preferred approach to organising and
  representing information
• Popular in adaptive systems
• Field dependent / independent need to see in
  context versus can start with details
• Holist / serialist top-down, starting with big
  picture versus bottom-up, needing well-defined,
  sequential steps
• Mainly adaptive navigation support

23
Individual TraitsCognitive Factors

• Other aspects of how your brain works that
  influence how you process information
• For exampleworking memory capacity, attention
  span
• Has a stable aspect
• But can change, e.g. due to being stressed or
  distracted

24
Individual Traits Learning Styles

• How people learn (cognitive styles related to
  learning)
• Mostly for Adaptive Educational Systems
• Mostly content adaptation
• For example
• Visualizer versus verbalizer some users prefer
  images ( remember better) while others prefer
  text (written or spoken)

25
Affective state

• Motivation, frustration, engagement, anxiety,
  happiness, anger,..
• Very changeable
• Separate lecture

26
Context of work

• User model may contain context features though
  these are not really user features
• For example
• Location important for mobile devices
• Device screen size, browser, bandwidth
• Time of day
• Social context alone, in company, at meeting
• Physical environment noise/light/temperature

27
Bayesian Networks

• More complicated models, with cause and effect
  relationships
• Calculates probabilities, and propagates them
  through a network

28
Bayesian Networks Example