Measuring Physiological Markers of Stress During Conversational Agent Interactions

Shreya Datar; Libby Ferland; Esther Foo; Michael Kotlyar; Brad Holschuh; Maria Gini; Martin Michalowski; Serguei Pakhomov

doi:10.1007/978-3-030-93080-6_18

Measuring Physiological Markers of Stress During Conversational Agent Interactions

Shreya Datar, Libby Ferland, Esther Foo, Michael Kotlyar, Brad Holschuh, Maria Gini, Martin Michalowski, Serguei Pakhomov

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Conversational agent (CA) technology is rapidly becoming ubiquitous. Understanding how CAs impact users on multiple levels, including physiology, thus becomes increasingly important. In this study, we examined the effects of a CA interaction on naive users’ physiological markers of stress i.e. heart rate (HR) and electrodermal activity (EDA). Participants (n = 21) prepared and executed a speech as part of a stressful interview, followed by a “Wizard-of-Oz” CA interaction. We expected the CA interactions to be mildly stressful. For a subset of participants with an initial resting period (n = 10), HR was elevated by 4.06 beats per minute (bpm) on average during the speech task, relative to the resting baseline. During the CA interaction however, HR was found to be 1.16 bpm lower on average. Moreover, HR and EDA values during the CA interaction were highly correlated with those during the resting period (Spearman’s rho: HR = 0.97, EDA = 0.96) with small differences (mean diff: HR = 1.16, EDA = 1.82). Contrary to initial expectations, users do not seem to exhibit a stress response during the CA interaction. We additionally performed similar analyses and compared our results with the Wearable Stress and Affect Detection (WESAD) dataset [1].

Original language	English (US)
Title of host publication	AI for Disease Surveillance and Pandemic Intelligence - Intelligent Disease Detection in Action
Editors	Arash Shaban-Nejad, Martin Michalowski, Simone Bianco
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	247-265
Number of pages	19
ISBN (Print)	9783030930790
DOIs	https://doi.org/10.1007/978-3-030-93080-6_18
State	Published - 2022
Event	5th International Workshop on Health Intelligence, W3PHAI 2021 held in conjection with 35th AAAI Conference on Artificial Intelligence, AAAI 2021 - Virtual, Online Duration: Feb 8 2021 → Feb 9 2021

Publication series

Name	Studies in Computational Intelligence
Volume	1013
ISSN (Print)	1860-949X
ISSN (Electronic)	1860-9503

Conference

Conference	5th International Workshop on Health Intelligence, W3PHAI 2021 held in conjection with 35th AAAI Conference on Artificial Intelligence, AAAI 2021
City	Virtual, Online
Period	2/8/21 → 2/9/21

Bibliographical note

Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Keywords

Conversational agent interactions
Physiological signal data
Stress detection
Wearable haptic garment
Wrist-worn device

Access

10.1007/978-3-030-93080-6_18

OpenUrl availability

Full text

Cite this

Datar, S., Ferland, L., Foo, E., Kotlyar, M., Holschuh, B., Gini, M., Michalowski, M., & Pakhomov, S. (2022). Measuring Physiological Markers of Stress During Conversational Agent Interactions. In A. Shaban-Nejad, M. Michalowski, & S. Bianco (Eds.), AI for Disease Surveillance and Pandemic Intelligence - Intelligent Disease Detection in Action (pp. 247-265). (Studies in Computational Intelligence; Vol. 1013). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-93080-6_18

Measuring Physiological Markers of Stress During Conversational Agent Interactions. / Datar, Shreya; Ferland, Libby; Foo, Esther et al.
AI for Disease Surveillance and Pandemic Intelligence - Intelligent Disease Detection in Action. ed. / Arash Shaban-Nejad; Martin Michalowski; Simone Bianco. Springer Science and Business Media Deutschland GmbH, 2022. p. 247-265 (Studies in Computational Intelligence; Vol. 1013).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Datar, S, Ferland, L, Foo, E, Kotlyar, M , Holschuh, B , Gini, M , Michalowski, M & Pakhomov, S 2022, Measuring Physiological Markers of Stress During Conversational Agent Interactions. in A Shaban-Nejad, M Michalowski & S Bianco (eds), AI for Disease Surveillance and Pandemic Intelligence - Intelligent Disease Detection in Action. Studies in Computational Intelligence, vol. 1013, Springer Science and Business Media Deutschland GmbH, pp. 247-265, 5th International Workshop on Health Intelligence, W3PHAI 2021 held in conjection with 35th AAAI Conference on Artificial Intelligence, AAAI 2021, Virtual, Online, 2/8/21. https://doi.org/10.1007/978-3-030-93080-6_18

Datar S, Ferland L, Foo E, Kotlyar M , Holschuh B , Gini M et al. Measuring Physiological Markers of Stress During Conversational Agent Interactions. In Shaban-Nejad A, Michalowski M, Bianco S, editors, AI for Disease Surveillance and Pandemic Intelligence - Intelligent Disease Detection in Action. Springer Science and Business Media Deutschland GmbH. 2022. p. 247-265. (Studies in Computational Intelligence). doi: 10.1007/978-3-030-93080-6_18

Datar, Shreya ; Ferland, Libby ; Foo, Esther et al. / Measuring Physiological Markers of Stress During Conversational Agent Interactions. AI for Disease Surveillance and Pandemic Intelligence - Intelligent Disease Detection in Action. editor / Arash Shaban-Nejad ; Martin Michalowski ; Simone Bianco. Springer Science and Business Media Deutschland GmbH, 2022. pp. 247-265 (Studies in Computational Intelligence).

@inproceedings{bb1d4696abbf4ae8baaeaac339a4a066,

title = "Measuring Physiological Markers of Stress During Conversational Agent Interactions",

abstract = "Conversational agent (CA) technology is rapidly becoming ubiquitous. Understanding how CAs impact users on multiple levels, including physiology, thus becomes increasingly important. In this study, we examined the effects of a CA interaction on naive users{\textquoteright} physiological markers of stress i.e. heart rate (HR) and electrodermal activity (EDA). Participants (n = 21) prepared and executed a speech as part of a stressful interview, followed by a “Wizard-of-Oz” CA interaction. We expected the CA interactions to be mildly stressful. For a subset of participants with an initial resting period (n = 10), HR was elevated by 4.06 beats per minute (bpm) on average during the speech task, relative to the resting baseline. During the CA interaction however, HR was found to be 1.16 bpm lower on average. Moreover, HR and EDA values during the CA interaction were highly correlated with those during the resting period (Spearman{\textquoteright}s rho: HR = 0.97, EDA = 0.96) with small differences (mean diff: HR = 1.16, EDA = 1.82). Contrary to initial expectations, users do not seem to exhibit a stress response during the CA interaction. We additionally performed similar analyses and compared our results with the Wearable Stress and Affect Detection (WESAD) dataset [1].",

keywords = "Conversational agent interactions, Physiological signal data, Stress detection, Wearable haptic garment, Wrist-worn device",

author = "Shreya Datar and Libby Ferland and Esther Foo and Michael Kotlyar and Brad Holschuh and Maria Gini and Martin Michalowski and Serguei Pakhomov",

note = "Publisher Copyright: {\textcopyright} 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 5th International Workshop on Health Intelligence, W3PHAI 2021 held in conjection with 35th AAAI Conference on Artificial Intelligence, AAAI 2021 ; Conference date: 08-02-2021 Through 09-02-2021",

year = "2022",

doi = "10.1007/978-3-030-93080-6_18",

language = "English (US)",

isbn = "9783030930790",

series = "Studies in Computational Intelligence",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "247--265",

editor = "Arash Shaban-Nejad and Martin Michalowski and Simone Bianco",

booktitle = "AI for Disease Surveillance and Pandemic Intelligence - Intelligent Disease Detection in Action",

address = "Germany",

}

TY - GEN

T1 - Measuring Physiological Markers of Stress During Conversational Agent Interactions

AU - Datar, Shreya

AU - Ferland, Libby

AU - Foo, Esther

AU - Kotlyar, Michael

AU - Holschuh, Brad

AU - Gini, Maria

AU - Michalowski, Martin

AU - Pakhomov, Serguei

PY - 2022

Y1 - 2022

N2 - Conversational agent (CA) technology is rapidly becoming ubiquitous. Understanding how CAs impact users on multiple levels, including physiology, thus becomes increasingly important. In this study, we examined the effects of a CA interaction on naive users’ physiological markers of stress i.e. heart rate (HR) and electrodermal activity (EDA). Participants (n = 21) prepared and executed a speech as part of a stressful interview, followed by a “Wizard-of-Oz” CA interaction. We expected the CA interactions to be mildly stressful. For a subset of participants with an initial resting period (n = 10), HR was elevated by 4.06 beats per minute (bpm) on average during the speech task, relative to the resting baseline. During the CA interaction however, HR was found to be 1.16 bpm lower on average. Moreover, HR and EDA values during the CA interaction were highly correlated with those during the resting period (Spearman’s rho: HR = 0.97, EDA = 0.96) with small differences (mean diff: HR = 1.16, EDA = 1.82). Contrary to initial expectations, users do not seem to exhibit a stress response during the CA interaction. We additionally performed similar analyses and compared our results with the Wearable Stress and Affect Detection (WESAD) dataset [1].

AB - Conversational agent (CA) technology is rapidly becoming ubiquitous. Understanding how CAs impact users on multiple levels, including physiology, thus becomes increasingly important. In this study, we examined the effects of a CA interaction on naive users’ physiological markers of stress i.e. heart rate (HR) and electrodermal activity (EDA). Participants (n = 21) prepared and executed a speech as part of a stressful interview, followed by a “Wizard-of-Oz” CA interaction. We expected the CA interactions to be mildly stressful. For a subset of participants with an initial resting period (n = 10), HR was elevated by 4.06 beats per minute (bpm) on average during the speech task, relative to the resting baseline. During the CA interaction however, HR was found to be 1.16 bpm lower on average. Moreover, HR and EDA values during the CA interaction were highly correlated with those during the resting period (Spearman’s rho: HR = 0.97, EDA = 0.96) with small differences (mean diff: HR = 1.16, EDA = 1.82). Contrary to initial expectations, users do not seem to exhibit a stress response during the CA interaction. We additionally performed similar analyses and compared our results with the Wearable Stress and Affect Detection (WESAD) dataset [1].

KW - Conversational agent interactions

KW - Physiological signal data

KW - Stress detection

KW - Wearable haptic garment

KW - Wrist-worn device

UR - http://www.scopus.com/inward/record.url?scp=85127058592&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85127058592&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-93080-6_18

DO - 10.1007/978-3-030-93080-6_18

M3 - Conference contribution

AN - SCOPUS:85127058592

SN - 9783030930790

T3 - Studies in Computational Intelligence

SP - 247

EP - 265

BT - AI for Disease Surveillance and Pandemic Intelligence - Intelligent Disease Detection in Action

A2 - Shaban-Nejad, Arash

A2 - Michalowski, Martin

A2 - Bianco, Simone

PB - Springer Science and Business Media Deutschland GmbH

T2 - 5th International Workshop on Health Intelligence, W3PHAI 2021 held in conjection with 35th AAAI Conference on Artificial Intelligence, AAAI 2021

Y2 - 8 February 2021 through 9 February 2021

ER -