Meta-Cognition, Motivation, and Affect

PSY504Spring term, 2011

April 6, 2011

Affect Detection

• What is a user’s affective state at a specific moment?

First note

• It can be done

• You have read a few examples of this– And the D’Mello & Calvo paper cites dozens more

examples

• More sources of data on affect leads to better detection, but it can be done reasonably well even with single data channels

Today…

• We will focus on the pragmatics of affect detection, rather than the technical details of building sensors, data processing, or detection algorithms

• Though please feel free to bring these types of issues up wherever they seem relevant

What are the requirements for useful detection for education?

• Example: Must be usable in real-time

What are the requirements for useful detection for education?

• Must be usable in real-time• Must work with no human intervention

– E.g. must be able to automate segmentation• Must be usable with real student data• Must be generalizable to population and situation of interest• Must be cost-effective• Users must be willing to comply• Breakage must be within affordable limitations• Privacy concerns when researchers use data• Must involve educationally-relevant affect

How would we establish each of these?

• Must be usable in real-time• Must work with no human intervention

– E.g. must be able to automate segmentation• Must be usable with real student data• Must be generalizable to population and situation of interest• Must be cost-effective• Users must be willing to comply• Breakage must be within affordable limitations• Privacy concerns when researchers use data• Must involve educationally-relevant affect

Information Used

• In detecting affect, researchers have used– Brainwaves (EEG)– Physiological Response (GSR, EKG, Eye movement)– Tone of voice– Facial expression– Posture/butt sensor– Mouse movements/keystrokes– Interaction features– Dialogue features– Contextual cues

What educational settings/conditions is each type of sensor feasible for?

• In detecting affect, researchers have used– Brainwaves (EEG)– Physiological Response (GSR, EKG, Eye movement)– Tone of voice– Facial expression– Posture/butt sensor– Mouse movements/keystrokes– Interaction features– Dialogue features– Contextual cues

Ground truth

Ground truth

• Expert judges• Self-report (in the moment, voluntary)• Self-report (in the moment, interruption)• Self-report (retrospective)

• Advantages/Disadvantages of each? Specific challenges?

Ground truth

• Periodic ratings of pre-defined time windows?

• Or report of onset of noticeable affective states?

• Advantages? Disadvantages? Specific challenges?

Ground truth

• For self-report, should you ask about – specific affective states – valence/arousal– valence

• Advantages? Disadvantages? Specific challenges?

Applications

• Within educational practice, what are some key applications of detecting a person’s affect?

• Within education research, what are some key applications of detecting a person’s affect?

Beyond 1-to-1 interactions

• Historically, affect detection has been applied to data from a single student working 1-on-1 with a computer

• Where else might affection detection be useful in education (or education research)?

Settings of use

• Almost all research on affect detection has been conducted in laboratory settings

• What are the limitations of this?

• What are the challenges in conducting affect detection research in ecologically valid settings?

• What can we learn from affect detection and corresponding affect research in laboratory settings, that is still useful for changing educational practice?

Next Class (APRIL 11)

• Affect and Achievement Goals

• Readings• Elliot, A.J., McGregor, H.A. (1998) Test Anxiety and the

Hierarchical Model of Approach and Avoidance Achievement Motivation. Journal of Personality and Social Psychology, 76 (4), 628-644.

• Pekrun, R., Elliot, A.J., Maier, M.A. (2006) Achievement Goals and Discrete Achievement Emotions: A Theoretical Model and Prospective Test. Journal of Educational Psychology, 98 (3), 583-597.