Developing Personas based on Physiological Measures

Vanessa Georges, François Courtemanche, Marc Fredette, Pierre-Majorique Léger

and Sylvain Sénécal

Tech3Lab, HEC Montréal, Montréal, Canada

Keywords: User Experience, Physiological Measures, Interaction Design, Personas.

Abstract: The objective of this paper is to propose a novel approach for the creation of user personas using common

patterns in psychophysiological signals. We illustrate the persona creation process through a case example.

Using this method, we were able to identify 4 distinct subgroups of varying experience and satisfaction levels.

This novel approach illustrates the potential of physiological measures in the identification of various user

clusters, based on one or more experiential aspect, as these signals can provide information as to what users

are experiencing during the interaction without interference. This should be useful for user experience

researchers, practitioners and designers alike to build more accurate user profiles, especially in the context of

large scale public installations and immersive experiences.

1 INTRODUCTION

In the context of design and user experience (UX)

practice and research, personas represent a group of

target users which share common behaviours, goals,

wants, needs, and frustrations when using a product

(Cooper, 1999).

However, according to McGinn et al., a common

issue with the development of personas is that they

are often not based directly on user data (McGinn and

Kotamraju, 2008). Expanding on this idea, Tu et al.

state that self-reported data from surveys, interviews

and user observation are not only disconnected from

user behaviour, but also weakly reflect users actual

use of a product (Tu et al., 2010).

To meet these challenges, researchers have

concentrated their efforts on finding new ways to

create personas based directly on user data. For

example, Zhang et al., have attempted to create

personas based on user behavior using clickstreams.

Others, like Tu et al. have tried to create data driven

personas using a multi-method approach, with both

qualitative data (i.e. observation and interview) and

quantitative data (i.e. cluster analysis) (Tu et al.,

2010). In the context of immersive interaction

environment, Loke et al. have proposed movement-

oriented personas and scenarios for representing

multiple users (Loke et al., 2005).

Adding to this body of work, our research

objective is therefore to propose a novel approach for

the creation of user personas that focuses on the

experiential dimension of human-computer

interaction (HCI), as we believe that the addition of

physiological measures in persona creation could

help improve on some of the well-documented

problems with persona use in HCI and UX design

practice. This should be particularly useful in the

context of interactive and immersive environments,

or any other circumstances where it may be difficult

to accurately observe and assess user experience.

An experiment was conducted to provide an

illustration of the potential of this novel method.

2 METHOD

Creating personas which reflect the unpredictable

changes in users can be challenging (Zhang et al.,

2016). To do so, we propose a novel approach: using

physiological measures in the development of user

personas. The proposed approach is based on the

analysis of users’ physiological signals, recorded

during their interaction. Our aim was to answer the

following question: is it possible to identify

subgroups of individuals with similar

psychophysiological states over time? In other words,

are we able distinguish groups of individuals with

similar emotional responses, experienced during the

same period of time, in a given experience?

Georges, V., Courtemanche, F., Fredette, M., Léger, P-M. and Sénécal, S.

Developing Personas based on Physiological Measures.

DOI: 10.5220/0006963201310136

In Proceedings of the 5th International Conference on Physiological Computing Systems (PhyCS 2018), pages 131-136

ISBN: 978-989-758-329-2

131

Physiological signals, such as electrodermal

activity, is indicative of emotional response and was

selected to provide valuable insights as to what users

are experiencing during the interaction, without

interference (Roto et al., 2009). To account for the

longitudinal (i.e. temporal) aspect of the experiment,

a curve clustering approach (Abraham et al., 2003)

was used to establish the existence of several

configurations of psychophysiological experiences.

Although Cugliari et al. used the technique to

identifying super consumers based on electricity

consumption, the curve classification approach has

never been used to our knowledge in the context of

psychophysiological data (Cugliari et al., 2015). An

experiment was conducted to provide an illustration

of the potential of this novel method.

2.1 Experimental Validation

An in-situ data collection was conducted to identify

and develop various user personas based personas

common patterns in psychophysiological signals

using curve clustering. The goal of the experiment

was to gather a large physiological data set

representative of different user experiences to assess

the accuracy of the proposed method.

2.1.1 Setting

The experience consisted of an interactive

multimedia installation in the forest, developed by a

company specialized in the creation of large scale

interactive experiences.

The location of the data collection was selected

among the installations already in operation and open

to the public. This gave us the opportunity to work

with an interesting physiological dataset without

having to build the interactive environment ourselves,

and to have access to a large and diverse population.

Figure 1: An image of the illuminated nocturnal trail.

The experience, which consisted of a one hour

walk, occurred entirely outdoors, in a forest in North

America, on a levelled dirt trail. The trail included

hills of moderate slopes. Access to 1.5-km

illuminated multisensory nocturnal trail (see figure 1,

2 and 3) was done via a chairlift. The trail weaves its

way down the hill through the woods, crosses streams

and clearings.

Figure 2: An image of the illuminated nocturnal trail.

Visitors also wore an interactive amulet which

changed color as they progressed through the trail.

Ambient sounds and an original soundtrack also

served to enhance visitors’ experience. Participants

Figure 3: The figure above shows an overview of the experience. Participants were fitted at the base camp located at the base

of the chairlift (zone 1). Zones 1, 4, 5, 6, 7, 10 and 11 were contemplative; zones 2, 3 and 9 were immersive; zone 8 was

interactive and zone 12 was participative.

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

132

were on the pathway an average of one hour, with a

maximum of 300 people admitted onto the course

every half-hour. Data was collected during the two

last weekends of September and the first weekend of

October. After completion of the experiment,

participants returned to base camp to remove the

physiological sensors and complete the post-

experiment questionnaire.

Figure 3 shows the configuration of the trail and

its various zones, including: three immersive zones,

seven contemplative zones, one interactive and one

participative zones.

2.1.2 Participants

For this experiment, a total of 36 participants, which

included 22 females, between the ages of 14 and 65

were recruited on site. Due to the considerable

difficulty of physiological data collection in natural

settings, data from 26 participants had to be discarded

from the analyses. Most data loss was due to signal

artefact and manipulation errors. Physiological data

from 10 participants were used from which 7 were

female. Participants were pre-screened for

neurological and psychiatric diagnoses. Data from all

36 participants were used in the questionnaire

analyses and curve clustering; data from 9

participants were rejected due to missing values (see

Section 2.2).

The total experiment duration was of one and a

half hours, which included the installation and

removal of the sensors, and beginning and end

questionnaires. A compensation corresponding to the

ticket price (i.e. 31.96 CAD) was given to each

participant upon completion of the experiment.

2.1.3 Physiological Signals and Equipment

A Bitalino (r)evolution Freestyle Kit BT (Lisbon,

Portugal) was used to record participants’

electrodermal activity (EDA). EDA was recorded

using two electrodes placed on the palm of the non-

dominant hand (see figure 4). An accelerometer was

also used to record motion.

EDA measures the activity of the eccrine sweat

glands and has been shown to be correlated to arousal.

It can also be used to measure emotions (Boucsein,

2012) during system interactions (figure 4). Valence

is used to contrast states of pleasure (e.g., happy) and

displeasure (e.g., angry), and arousal to contrast states

of low arousal (e.g., calm) and high arousal (e.g.,

surprise). The sensor enclosure box was placed on

participants’ left arm and secured with an exercise

band. EDA was recorded with a sampling rate of

100Hz.

Figure 4: The amulets and physiological sensors.

EDA was

recorded using two electrodes placed on the palm of the

non-dominant hand.

2.1.4 Self-reported Data and Questionnaires

In addition to socio-demographic questions, a 9-point

SAM scale was used to measure self-reported arousal

and valence of participants before and after the

experience. Questions on customer expectations and

prior immersive experiences were also included in the

questionnaire.

Customer satisfaction was assessed in two

different ways. Participants were asked to report their

own satisfaction regarding their overall experience. A

satisfaction score was also calculated using the ACSI

score, the American Consumer Satisfaction Index

(Fornell et al., 1996).

2.2 Preliminary Findings and Results

Physiological data from 10 participants were used for

the curve clustering. Clustering was done using the

galvanic skin response (GSR) based on the number of

peaks per minute in each station, which better

illustrates the psychophysiological and emotional

responses (Garrett, 2010) of participants, as opposed

to EDA means. Values were standardized at

participant level.

Using this method, we were able to distinguish 4

groups of participants who experienced similar levels

of arousal during the same station. Figure 5 illustrates

variation of the GSR means of the 4 clusters

throughout the 12 stations.

In the context of this experiment, Cluster 1

appears to describe people who are engaged at the

start of the experience, but whose interest

significantly drops off towards the end. On the other

hand, cluster 2 seems to feature participants who start

and end the experience on high notes, but experience

little activation throughout. Cluster 3 seems to

experience a more gradual drop in engagement. As

Developing Personas based on Physiological Measures

133

for cluster 4 his experience, despite its ups and downs,

seems to remain more constant.

Figure 5: Curves of the 10 participants based on

average GSR values per station.

To verify the robustness of our analysis,

benchmarking of physiological data to the self-

reported questionnaire data was done. In this step of

the process, 2 participants were rejected due to

missing values in the self-reported questionnaire data.

These clusters, identified using EDA, were

corroborated by questionnaire data. In other words,

had we used the sociodemographic and psychometric

variables of the questionnaires to create these

clusters, 7 out of 8 participants would still have been

assigned to the same subgroup.

2.3 Cluster Creation

Given this high success rate, our next step was to

attempt to classify the remaining participants. In other

words, we used the self-reported questionnaire data to

infer into which cluster each of the remaining 26

participants would belong to (see table 1).

Sociodemographic and psychometric variables were

used to reproduce the above clustering results by

minimizing the Euclidean distance to the centroids of

the above clusters (figure 5).

Table 1: The four clusters by mean of self-reported data.

Cluster cl1 cl2 cl3 cl4

valence_

egin 5 4,14 1,5 3,2

arousal_en

3,75 2,14 5,6 5,2

valence_en

2,5 5,43 2,2 3,93

appre_amulet 7,5 4,07 6,9 6,9

satis_ACSI 67,6 43,39 69,64 69,64

education 3 4 4,8 4,8

arousal_

egin 6,25 4,71 5,7 5,7

appre_story 6,5 4,93 6,95 6,95

satis_overall 5,75 4,71 6,45 6,45

N 4 7 10 15

Below, a breakdown of the four subgroups

describing clusters by mean of the questionnaire

variables.

Cluster 1:

 Highest reported valence at the beginning

 Highest reported arousal at the beginning

 Significant drop in arousal from beginning

to end

 Lowest level of education

Cluster 2:

 Lowest reported arousal at the end

 Lowest reported appreciation of the amulet

 Lowest reported appreciation of the narrative

 Highest reported valence at the end

 Lowest calculated ACSI satisfaction score

 Lowest reported satisfaction level overall

 Significant drop in arousal from beginning

to end

Cluster 3:

 Lowest reported valence at the beginning

 Highest reported arousal at the end

 Highest reported education level

 Lowest reported valence at the end

 Highest level of education

Cluster 4:

 Highest reported appreciation of the story

 Highest calculated ACSI satisfaction score

 Highest reported appreciation of the amulet

 Lowest reported level of education

 Lowest reported arousal at the beginning

 Highest reported satisfaction overall level

 Highest level of education

As previously mentioned, personas describe groups

of target users which share common behaviours,

goals, wants, needs, and frustrations when using a

product (Cooper, 1999). While these user profiles do

not yet represent complete personas, the inclusion of

physiological measures in combination with

traditional methods (i.e. interviews, questionnaires,

etc.), can help UX designers to better understand the

needs, behaviour and frustrations of users during

interactive experiences.

With the addition of qualitative data, this method

should help designers by providing an additional level

of detail; taking these above clusters from groups of

individuals with similar emotional responses

experienced during the same period of time, to user

based personas.

2.4 Main Conclusions

Using physiological measures as the statistical

starting point, we were able to identify four subgroups

using a curve clustering method. Although we chose

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

134

a difficult data collection setting due to partnership

constraints, we were able to identify various personas

that can help designers improve this interactive trail.

In the context of this experiment, we can observe

a significant drop in self-reported arousal from the

beginning to the end of the experience end can also

be indicative of a lower overall satisfaction level.

Cluster 4 was the only subgroup with a self-reported

arousal increase from the beginning to the end of the

experience. This subgroup also has the highest self-

reported and calculated overall satisfaction. This

cannot be said of valence.

One of the main challenge of the data collection

was the environment itself. The experience occurred

entirely outdoors in a forest, on a leveled but non-

asphalted trail. This caused movement artefact.

Furthermore, the effects of ambient and skin

temperatures fluctuations on EDA have long been

proven (Edelberg, 1972). These environmental

limitations explain in large part the small number of

participants included in the physiological data

analysis, as many subjects were rejected due to poor

quality signal. However, using both qualitative and

quantitative not only added depth to our analysis, but

also allowed us to recover important participant data.

Therefore, this method should be even more efficient

under better conditions.

3 CONCLUSION

This novel approach illustrates the potential of

physiological measures in the identification of

personas based on one or more experiential aspect.

Although neither personas nor physiological

measures are new to HCI or UX, the combination of

the two could help user profiling by bringing groups

of archetypal users to life, in order to support user-

centred design practice. This novel approach also

responds to a need for more data-driven personas,

based directly on user data, as we can see even here

the discrepancies between experienced and self-

reported arousal. This method should be particularly

useful to HCI researchers, practitioners and

designers, especially in the context of interactive and

immersive environments, or any other circumstances

where it may be difficult to accurately observe and

assess user experience.

4 FUTURE WORKS

The next step is to further develop this method in a

more controlled environment, for example a business

conference or concert, which will allow us to collect

quality data on a much bigger sample size. This will

also enable us to include other physiological signals,

such as heart rate and mobile eyetracking.

ACKNOWLEDGEMENTS

Authors want to thank the research assistants who

administered the study. This work was supported by

the Natural Sciences and Engineering Research

Council.

REFERENCES

McGinn, Jennifer Jen, and Nalini Kotamraju. "Data-driven

persona development." Proceedings of the SIGCHI

Conference on Human Factors in Computing Systems.

ACM, 2008.

Edelberg, Robert. "Electrical activity of the skin: Its

measurement and uses in psychophysiology."

Handbook of psychophysiology 12 (1972): 1011.

Boucsein, Wolfram. Electrodermal activity. Springer

Science & Business Media, 2012.

Garrett, Jesse James. Elements of user experience, the:

user-centered design for the web and beyond. Pearson

Education, 2010.

Cugliari, Jairo, Yannig Goude, Jean-Michel Poggi, Jairo

France, and Jean-Michel France. "Classification de

courbes individuelles et prévision désagrégée de la

consommation électrique.", 2015.

Abraham, Christophe, Pierre-André Cornillon, E. R. I. C.

Matzner‐Løber, and Nicolas Molinari. "Unsupervised

curve clustering using B‐splines." Scandinavian

journal of statistics 30, no. 3 (2003): 581-595

Tu, Nan, Xiao Dong, Pei-Luen Patrick Rau, and Tao Zhang.

"Using cluster analysis in persona development." In

Supply Chain Management and Information Systems

(SCMIS), 2010 8th International Conference on, pp. 1-

5. IEEE, 2010.

Zhang, Xiang, Hans-Frederick Brown, and Anil Shankar.

"Data-driven Personas: Constructing Archetypal Users

with Clickstreams and User Telemetry." Proceedings of

the 2016 CHI Conference on Human Factors in

Computing Systems. ACM, 2016.

Loke, Lian, Toni Robertson, and Tim Mansfield. "Moving

bodies, social selves: movement-oriented personas and

scenarios." Proceedings of the 17th Australia

conference on Computer-Human Interaction: Citizens

Online: Considerations for Today and the Future.

Computer-Human Interaction Special Interest Group

(CHISIG) of Australia, 2005.

Roto, Virpi, Marianna Obrist, and Kaisa Väänänen-Vainio-

Mattila. "User experience evaluation methods in

academic and industrial contexts." Proceedings of the

Workshop UXEM. Vol. 9. 2009.

Developing Personas based on Physiological Measures

135

Fornell, Claes, Michael D. Johnson, Eugene W. Anderson,

Jaesung Cha, and Barbara Everitt Bryant. "The

American customer satisfaction index: nature, purpose,

and findings." the Journal of Marketing (1996): 7-18.

Cooper, Alan. The inmates are running the asylum: [Why

high-tech products drive us crazy and how to restore the

sanity]. Indianapolis: Sams, 2004.

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

136