Summary of

1
Summary of 3rd Workshop on Affective and
Attitude User Modeling

• Assessing and Adapting to User Attitudes and
  Affect Why, When and How?
• Co-Chairs
• Cristina Conati, Eva Hudlicka, Christine Lisetti

2
Objectives

• Explore core issues in expanding user models to
  include information of affect and attitudes
• Detection
• Modeling
• Adaptation
• Validation and Evaluation
• ... across a range of applications

3
Focus Questions

1. Can we begin to identify a taxonomy of
   circumstances requiring affective user modeling?
2. What are existing methods of constructing
   affective user models? (detection/modeling)
3. Can we define guidelines for model use?
   (adaptation)
4. What are methods and issues in validation and
   evaluation?

4
Workshop Organization

• 8 long papers in 3 Sessions
• 5 posters in Poster Session
• Small Group Discussion and Presentation
• Panel (Fiorella de Rosis, Anthony Jameson, Roz
  Picard, Amy Kruse)

5
Identify a Taxonomy of Circumstances Requiring
Affective User Modeling?

• Possible Applications
• Training and Tutoring
• Dialogue Systems
• CSCW
• Telehealth

Chainsaw
??

• Critical Factors
• Affect central to task (e.g. psychotherapy)
• Affect influences user performance (e.g.,
  learning)
• Degree to which adaptation is possible
• Placement on spectrum must be empirically tested

6
Can We Define Guidelines for Model Use When and
How to Adapt?

• Factors
• Context
• Users goals and traits
• Target of the adaptation
• User affective state
• Consequences of affective states (e.g.,
  attention)

• Critical issues
• User control
• Empirical evaluations of adaptation strategies

7
What are Existing Methods of Constructing
Affective User Models?

• Detection
• Biometric Sensors Effectiveness varies along
  various dimensions
• Diagnosticity (GSR good for arousal)
• Intrusiveness (depends on application, wireless
  is good)
• Reliability (BVP and Respiration low)
• NL and speech (Emotionally tagged grammars)

• Modeling
• Bayesian and Dynamic Belief Networks
• Finite State Machine
• Other methods analized in small group discussion

8
What are Methods and Issues for Validation and
Evaluation?

• EVALUATION, DAMN IT!

9
What are Methods and Issues for Validation and
Evaluation?

• Evaluate the detection/modeling directly
• Challenges
• Ephemeral
• Idiosyncratic
• Not always observable
• Possible Solution Multiple Measures (e.g.
  observations self-reports)

• Evaluate the adaptation built on the model for
• User satisfaction
• Task performance

10
Conclusions

• Prototype systems exist which can sense
  frustration / stress and provide adaptive
  response (e.g., empathy)
• Variety of promising detection/modeling methods
  exist
• Critical Needs/Issues
• Corpora of Labelled Data
• Gold standards for validating measures
• Clear definition of constructs
• Privacy
• Understand level of modeling resolution required
  do not overmodel!

11
For moresee workshop Web pagewww.cs.ubc.ca/cona
ti/um03-affect/affectum.html

• Thanks to workshop and panel
• partecipants!