Interaction-based affect detectors outperformed posture-based detectors, with boredom detected best and posture detectors near chance¶

Claim · All claims
Evidence · 1 study (2 entries) · associational r2 · q2
Currency · newest study 2018 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Interaction-based detectors of affect in TC3Sim showed highly variable performance, with boredom best (Kappa = 0.469, A' = 0.848) and confusion barely above chance (Kappa = 0.056, A' = 0.552). → DeFalco 2018 q2 i? Posture-based detectors performed only slightly better than chance, with surprise worse than chance and boredom best (Kappa = 0.109, A' = 0.528). → DeFalco 2018 (2)

Evidence¶

DeFalco 2018¶

DeFalco, J. A., Rowe, J. P., Paquette, L., Georgoulas-Sherry, V., Brawner, K., Mott, B. W., Baker, R. S., & Lester, J. C. (2018). Detecting and Addressing Frustration in a Serious Game for Military Training. Int J Artif Intell Educ. https://doi.org/10.1007/s40593-017-0152-1

q2 · i? · causal · r2

Detector development study with 119 USMA cadets using TC3Sim; interaction-based detectors were validated with 10-fold participant-level cross-validation against BROMP field observations. The boredom detector achieved "Kappa = 0.469, A'= 0.848" while confusion performed barely above chance.

"Performance of our interaction-based detectors was highly variable across affective states. The interaction-based detector of boredom achieved the highest performance (Kappa = 0.469, A'= 0.848) while some of the other detectors achieved very low performance."

DeFalco 2018 (2)¶

DeFalco, J. A., Rowe, J. P., Paquette, L., Georgoulas-Sherry, V., Brawner, K., Mott, B. W., Baker, R. S., & Lester, J. C. (2018). Detecting and Addressing Frustration in a Serious Game for Military Training. Int J Artif Intell Educ. https://doi.org/10.1007/s40593-017-0152-1

q2 · i? · associational · r2

Kinect posture-based detectors built from 73 posture features over the same BROMP-labeled corpus; the best posture detector (boredom, logistic regression) achieved Kappa = 0.109, A' = 0.528, and the frustration SVM achieved Kappa = 0.061, A' = 0.518.

"In general, posture-based detectors performed only slightly better than chance, with the exception of the surprise detector, which actually performed worse than chance."

Discussion¶

Related Claims¶

• During novices' first programming lesson, engagement, confusion, frustration, boredom, and curiosity are the dominant affective states, collectively accounting for approximately 73% of affect judgments — related