=Paper=
{{Paper
|id=Vol-2885/paper3
|storemode=property
|title=Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
|pdfUrl=https://ceur-ws.org/Vol-2885/paper3.pdf
|volume=Vol-2885
|authors=Charles Nutrokpor,Akon O. Ekpezu,Abigail Wiafe,Isaac Wiafe
|dblpUrl=https://dblp.org/rec/conf/persuasive/NutrokporEWW21
}}
==Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps==
https://ceur-ws.org/Vol-2885/paper3.pdf
Exploring the Impact of Persuasive System Features on
User Sentiments in Health and Fitness Apps
Charles Nutrokpor1 [0000- 0002-9155-4556], Akon O. Ekpezu2 [0000-0002-9502-1052], Abigail
Wiafe3 [0000 -0001-6019-6074], and Isaac Wiafe 2 [0000-0003-1149-3309]
1
Department of Computer Science, University of Ghana, Accra, Ghana
2Faculty of Information Technology and Electrical Engineering, University of Oulu, Finland
3School of Computing University of Eastern Finland, Joensuu, Finland
iwiafe@ug.edu.gh
Abstract. Although behavioural change support systems have proven to be ef-
fective in changing user’s behaviour, the need to design effective persuasive sys-
tems that optimize persuasive experiences of users continue to remain a chal-
lenge. This study seeks to contribute to existing literature that aims at addressing
this challenge. Using stratified random sampling technique, 23 health and fitness
iOS and Android apps were selected. User reviews of each app were downloaded
and compared with the corresponding persuasive systems features using cluster
analysis. The findings demonstrated that more system features do not produce
higher positive sentiment. It was also observed that apps with more social support
features were associated with higher frequencies of fear, sadness and anger re-
lated sentiments.
Keywords: Persuasive and Sentiments, Health Behaviour Change Support Sys-
tem, Sentiment analysis, Mobile Health, Health and fitness apps
1 Introduction
Since the introduction of the Persuasive Systems Design (PSD) framework [1] several
studies have attempted to investigate the efficacy of the 28 suggested persuasive fea-
tures in different domains [2]–[4]. The PSD framework proposed system features that
are categorized into four main supports (i.e., Primary Task Support, Credibility Sup-
port, Dialogue Support, and Social Support) for changing behaviour. These features are
the fundamental system requirements of a behaviour change support system (BCSS),
and although it is not mandatory for all the features to be present for a system to be
considered as a persuasive [1] there is the need for some representation of these features
to be present. A key challenge in BCSSs research is to determine how to select the most
relevant persuasive features to increase the persuasive experience of users [5]. Accord-
ingly, several studies [5], [6] have proposed methods and frameworks for selecting per-
suasive systems features to optimize user persuasive experience. Yet, these methods do
not adequately provide information on how system features can be selected. To address
this challenge, this study seeks to contribute by exploring how various persuasive sys-
tem features trigger specific sentiments or emotions in users.
Copyright © 2021 for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
26 Ninth International Workshop on Behavior Change Support Systems (BCSS 2021):
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
It is emphasized that, although some studies have attempted to assess or evaluate the
relationship between persuasive features and their impacts on users [2], [4], [7], [8], to
our knowledge none have investigated the effects of persuasive systems design features
on user sentiments. However, understanding how system features trigger specific sen-
timents provide pertinent information that may aid the selection of effective and effi-
cient persuasive features. This is because there is enough evidence that emotions or
sentiments moderate human behaviour and thus impacts persuasion [9].
Specifically, this study assessed 23 selected health and fitness mobile applications
in the Android and iOS markets and explored the relationship between users’ senti-
ments and app features. Next is a discussion on related literature. This is followed by a
description of how the study was conducted. The findings and implications of the study
are presented before conclusions are drawn.
2 Background Literature
2.1 Persuasive Systems Design Features and Related Studies
The intention to change one's behaviour using technology depends on three major fac-
tors, the designer, the distributor and the user [10]. Considering that the main preroga-
tive of BCSSs is to alter behaviour, it is incumbent for designers to ensure that they
employ techniques that facilitate persuasion by optimizing the use of persuasive fea-
tures. Yet, studies have shown that persuasive software features are not mostly consid-
ered by designers during the design stage [11] and also most persuasive designers em-
ploy ad hoc design methods [12]. Persuasive systems design features provide a means
for designers to enhance the content and or functionalities of persuasive software. The
28 PSD features are primary task support (reduction, tunnelling, tailoring, personaliza-
tion, self-monitoring, simulation, and rehearsal), dialogue support (praise, rewards, re-
minders, suggestion, similarity, liking, and social role), system credibility support
(trustworthiness, expertise, surface credibility, real-world feel, authority, third-part en-
dorsements, and verifiability), and social support (social learning, social comparison,
normative influence, social facilitation, cooperation, competition, and recognition) [1].
It has been argued that a good understanding of these features and their impact on spe-
cific persuasive activities provide the needed information that facilitates the design of
effective persuasive systems [13]. Nonetheless, it is a challenge to identify specific and
exact features that enhances persuasion. This challenge is a result of the complex nature
of human attitude and behaviour, and it was inherited from traditional methods for
changing human behaviour.
That notwithstanding, several studies have attempted to understand the relationship
between persuasive system features as proposed by Oinas-Kukkonen and Harjumaa [1]
and the possible impacts it has on persuasion [13], [14], usability, credibility and con-
tinuous usage [2], [7]. These studies have however produced relatively conflicting re-
sults. For instance, it has been argued that the presence of persuasive system features
in Health Behaviour Change Support System (HBCSS) does not necessitate the suffi-
ciency and or efficiency of the system, rather more attention should be given to design-
ing and implementing systems that are captivating and attractive to users [13]. Others
Ninth International Workshop on Behavior Change Support Systems (BCSS 2021): 27
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
have argued that perceived effectiveness, availability, and credibility (trust, reliability,
etc.) of a system has a direct impact on user intention to continuous use of BCSS [13],
[14]. Accordingly, there is a need for further investigations on how these features im-
pact persuasive design from a different perspective.
2.2 Sentiment and Persuasion
Due to complexities in understanding human attitude, behaviour and the limitations of
using questionnaires to collect and investigate perceptions, it is more appropriate to
adopt other self-reporting methods that do not involve questionnaires to study human
behaviour. Thus, recent studies have adopted sentiment analysis for investigating hu-
man emotions and behaviour. Sentiment analysis provides a better option for studying
user perceptions. Mostly, users express their opinions on applications or products to
demonstrate their level of satisfaction and these opinions provide rich information for
investigations. In recent times, the web has become a viable space where individuals
express their opinions. Internet reviews have become a relevant part of decision-making
processes for individuals and industries. Particularly, user feedback is a fundamental
variable for purchase decisions, and it provides relevant information for determining
the satisfaction levels and emotions of customers.
Considering BCSS designs, existing evidence confirms that there is a relationship
between sentiments and persuasion [15]. Persuasion is a communication activity which
present arguments to motivate or change the cognitive state of the listener [16]. Thus,
persuasion techniques exert influences on the thoughts and behaviour of individuals,
and this induces sentiments. A change in an individual’s sentiment may affect behav-
iour and this has been demonstrated in how sentiments expedite decision making [17]–
[19]. Individuals rely on their emotions to make economic, political, social and personal
decisions. It is, however, evident that the extent of decision making based on emotions
can be biased: whether deducted from persuasive messages or incidental contextual
factors. This notion has been confirmed by Petty & Cacioppo [15] in the Elaboration
Likelihood Model (ELM) that explains the effect of emotions on attitude and judge-
ment.
In BCSS design, emotions play a crucial role in translating the effects of feeling from
computers (application) to humans [20]. Hence, emotions can influence a user’s ac-
ceptance of a BCSS. Incorporating emotional strategies into persuasive messages might
motivate a user towards achieving their persuasive goals. For example, evoking fear
can be a good means of alerting an individual of the risks of heart disease due to smok-
ing [21]. Yet, a critical observation of BCSS design literature demonstrates inadequate
investigations on the relationship between sentiments and persuasive features. Studies
have mainly focused on individual emotions such as fear [22], trust [23] and self-re-
flection [24]. It has been argued that positive emotions increase trust while negative
emotions decrease it [25]. Nonetheless, in BCSS, cognitive trust has a higher impact on
credibility and continuous use when compared to affective trust [23]: a decrease in cog-
nitive trust is directly proportional to a decrease in affective trust [25]. As argued ear-
lier, considering the implications of current literature, it is relevant to investigate or
28 Ninth International Workshop on Behavior Change Support Systems (BCSS 2021):
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
explore the relationship between sentiments and persuasive features. Accordingly, this
study sought to explore this relationship in health and fitness mobile applications.
2.3 Persuasive Mobile Health
Health and fitness application was adopted because of its popularity in recent times. It
has demonstrated to be effective in addressing several health-related issues. Conse-
quently, research on the use of persuasive features in health-related apps has gained
more attention [12]. Some researchers have argued that mobile health applications pre-
sent a better opportunity for addressing barriers to patient education [26] and disease
prevention [11], [24]. Mobile health apps are ubiquitous and pervasive, thus, more ac-
cessible when compared to traditional systems. More specifically, health apps on mo-
bile phones and smart devices have addressed challenges of infrequent usage of web-
based health intervention: smartphone users are more responsive to behaviour change
strategies available in mobile health and fitness apps [27]. It has been argued that alt-
hough there is no significant difference in mortality rates between users and non-users
of mobile health apps, mobile apps have reduced hospital admission rates and have also
improved health outcomes such as lower systolic blood pressure and medication com-
pliance significantly [26].
However, existing mobile health applications seek to promote healthier habits by
improving its technology [11] rather than paying attention to the fundamental persua-
sive principles that addresses consumer needs. Specifically, existing applications can
be improved by leveraging effective persuasive system features to provide effective
communication and persuasion. Considering this backdrop and the widespread use of
mobile health apps, this study adopted mobile health application as the domain of in-
vestigation.
3 Methodology
To ensure a compressive and rigorous review of sentiments and features of mobile
health apps, the study was conducted as follows: firstly, a sample of mobile apps cate-
gorized as “health and fitness” were selected from the Android and iOS stores. Each
app was assessed based on an approved selection criterion. The persuasive features and
the associated sentiments of the selected apps were extracted. The patterns in app design
features and related sentiments were explored to draw conclusions. Below is a detailed
discussion on how each stage of the investigation was conducted.
3.1 Datasets
The dataset for the study was acquired from the Kaggle datasets for iOS and Android.
The Kaggle datasets for Google Play store apps (https://www.kaggle.com/gau-
thamp10/google-playstore-apps) and Apple iOS app store
(https://www.kaggle.com/cmqub19/763k-ios-app-info) were downloaded (on Septem-
ber 14, 2020). The database consists of 735,593 and 4,175 applications classified as
health and fitness for Android and iOS respectively. The dataset was pre-processed and
Ninth International Workshop on Behavior Change Support Systems (BCSS 2021): 29
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
fields or data that were considered to be irrelevant for the study were excluded. Further,
apps that had less than 500 reviews were excluded. This was to ensure that all apps used
in the study have received an adequate number of reviews and ratings. Duplicate apps
including those that were present in both iOS and Android were removed. This reduced
the number of apps to 278. (i.e., 99 apps for iOS and 179 for Android).
For a population of 278 applications, a sample of 72 is needed to ensure a 10% mar-
gin of error at a 95% confidence level. A stratified sampling approach resulted in 28
samples for iOS and 44 for Android. Our motivation to use a stratified random sample
approach was to reduce biases and ensure that the findings of this study can be gener-
alized. Each application was downloaded and installed. After installation, applications
that were not in English, those that were for sale, no longer available or did not demon-
strate an intention of changing user behaviour were omitted. This resulted in 23 appli-
cations for the study. Figure 1 is a diagrammatic representation of the stages involved
in the selection process and Table 1 is a list of the selected apps used for the study.
The reviews and ratings for these applications were extracted using Python libraries
(i.e., beautifulsoup, selenium and JSON). Downloaded reviews for individual apps
ranged from 202 to 123,719. Each review consists of ratings, categorical_url, com-
pany_name, date, developerResponse, reviews/content, title, and isEdited.
Fig. 1. Procedure for selecting apps
30 Ninth International Workshop on Behavior Change Support Systems (BCSS 2021):
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
3.2 Persuasive Feature Extraction
Two members of the research team were tasked to extract the various persuasive fea-
tures of the selected apps. They used each app for one month simultaneously to assess
the apps and identify the various persuasive features employed in each app. To reduce
bias, reports from the two assessors were combined and disparities were addressed.
Similar to studies conducted by Lehto and Oinas-Kukkonen [13], features including
liking and similarity were not assessed. This is because they are relatively subjective,
ambiguous and dependent on the user. Although, it is challenging to assess surface
credibility and trustworthiness, in this study surface credibility was evaluated using
claims by [11]. Thus, the absence or minimal use of adverts and unnecessary pop-ups
was used to assess surface credibility whereas trustworthiness was evaluated by the
ability of the application to provide users with control of security/privacy settings.
Table 1 List of Applications used for the Study
App ID Name of Application App ID Name of Application
1 Ideal Weight 13 Step Counter - Calorie Counter
2 WalkingApp 14 Weight Loss Running by Verv
3 Step Counter 15 Abs Workout
4 Walking for Weight Loss 16 Pocket Yoga
5 Headspace 17 Dr.Greger's Daily Dozen
6 Calorie Counter by FatSecret 18 HidrateSpark Smart Bottle
7 Cycling - Bike Tracker 19 Jillian Michaels Fitness App
8 Running Distance Tracker + 20 PlayFitt
9 Daily Yoga - Yoga Fitness Plans 21 WaterLama Water Tracker
10 Pregnancy & BabyTracker 22 Six Pack in 30 Days
11 Workout Tracker & Gym Trainer 23 Plant Nanny
12 Water Drink Reminder
3.3 Sentiment Extraction and Analysis
The reviews and ratings for each selected app were extracted and pre-processed. Data
pre-processing is an essential part of sentiment analysis (i.e., Natural Language Pro-
cessing). It enables the stemming and elimination of redundant data such as stop words
and noise. Hence, stop words including prepositions, pronouns, special characters,
punctuation marks and numbers were eliminated from the dataset. Furthermore, to
avoid short words pollution and eliminate words that were not removed during stop
words removal, words with three characters and below (e.g., eat, run, got) were re-
moved. Two categories of sentiments were considered: the opinion sentiments consist-
ing of positive or negative and emotional sentiments consisting of five classes of emo-
tions namely liking, trust, anger, sadness and fear. These five classes were identified in
an initial exploration of the dataset that identified them as the main classes present in
the dataset. The five classes of emotions were categorized by synonyms and related
words. Due to mix of words relating to adjectives, nouns, adverbs and verbs that can be
found within the list of sentimental words, the wordnet database was used to find other
synonyms. See table 2 for the categorization of words for the classes used in this study.
Ninth International Workshop on Behavior Change Support Systems (BCSS 2021): 31
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
A sentiment intensity calculation was performed, here the total number of opinions
and emotional sentiments were analysed. To accumulate the exact sentiments extracted
from the reviews, sentiment extraction was conducted in two folds. The first fold used
a four-way approach for categorizing emotional sentiments; Classification of Reviews,
Frequency of Words, Extracting Sentimental Words and General Sentimental Group-
ing. The second fold used a three-way approach; calculating the percentage of the opin-
ion sentiments, categorizing the opinion sentiments into the five stated emotional sen-
timents and calculating the percentage of the total number of positive and negative sen-
timental words respectively.
The reviews and their corresponding ratings were grouped into positive and negative
words. Using a word extraction function, each app review was evaluated to determine
whether or not their ratings fell above or below three (3). Additionally, sentiment re-
trieval was performed using the frequency of words and the extraction of sentimental
words. Words from both the positive and negative lists were combined and a Frequency
Distribution function was used to output a dictionary of the most frequent words within
the list. This facilitated the identification of relevant words for each application. Each
word and its corresponding frequency distribution were placed into a data frame. The
sentiments were extracted from the data frames and analysed. The Valence Aware Dic-
tionary for Sentiment Reasoning (VADER) model was used as the sentiment analyser.
Words with compound exposure of 0.5 or -0.5 based on their polarity property of
VADER were combined to form the list of sentimental words with their positive and
negative sentimental intensity. The combined words were split into their respective pos-
itive and negative sentiments and the polarity of each app was calculated.
K-Means clustering approach was used to assess the relationship between persuasive
systems features and their respective sentiments. The Elbow method for selecting the
optimal k clusters produced 6 clusters as the optimal number of clusters. The dataset
was fitted on K-Means where n_clusters = 6 and random_state = 42. The predicted
outcome of the computation was retrieved and analysed.
4 Findings and Discussion
4.1 Characteristics of Selected Apps
Findings from the persuasive feature extraction demonstrated that no application used
all the 28 persuasive systems features. However, Primary Task support was dominant
in health and fitness applications. With regard to Dialogue support features, 18 out of
the 23 applications used Reminders and 20 used Suggestion. These were the two most
used Dialogue support feature. In most cases, applications that used reminders also used
suggestions. Praise (11), rewards (10) and social role (12) were averagely used. A no-
table observation in the evaluation of Credibility support features was that there was a
relationship between the presence of trustworthiness and surface credibility. Trustwor-
thiness was present in 22 out of the 23 applications evaluated whereas surface credibil-
ity was present in 21. Third-party endorsement (6) and authority (4) were sparingly
used. Overall, Social support features were the least adopted features. Social learning
32 Ninth International Workshop on Behavior Change Support Systems (BCSS 2021):
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
was observed in 16 applications and 10 used social facilitation. Normative influence
(9), cooperation (6), social comparison (5), recognition (4), and competition (2) were
barely used. Refer to table 3 for a complete list of persuasive systems features identified
in the 23 mobile health and fitness apps evaluated. These findings revealed that Primary
Task support features are dominant in mobile health apps and this confirms current
knowledge [11], [28]. Also, Social support features are sparingly used. Similar claims
have been made on a study that investigated persuasive system features of e-commerce
platforms [28].
4.2 Relationship between Sentiments and System Features
It was observed that applications including WalkingApp (app2), Step Counter (app3),
Headspace (app5), Calorie Counter by FatSecret (app6), Running Distance Tracker +
(app8), Daily Yoga (app9), Pregnancy & Baby Tracker (app10), HidrateSpark Smart
Bottle (app188), Jillian Michaels Fitness App (app19), Six Pack in 30 Days (app22)
and Plant Nanny (app23) were in one cluster (i.e., C1). See table 3 for a list of the
various apps and the corresponding clusters labelled as C1 to C6. This cluster set was
characterized by a high frequency of Primary task support features including reduction,
tunnelling, tailoring, personalization and self-monitoring. Simulation and rehearsals
were present, however, they had lower frequencies. With regard to dialogue support
features, praise, reminders, suggestion and social role were present with high frequen-
cies whilst rewards had a low frequency. For Credibility support, high frequencies were
observed for trustworthiness, expertise, surface credibility, real-world feel and verifia-
bility whilst authority and third-party endorsement had lower frequencies. All seven (7)
features within Social support were present in this cluster (i.e., C1), however, they were
marginally represented. Again, in terms of opinion sentiments, this group of mobile
applications had higher positive sentiment values except for Headspace (app5) which
record a low positive sentiment.
Walking for Weight Loss (app4), Workout Tracker & Gym Trainer (app11), Abs
workout (app15) formed a cluster (i.e., C2). This group of apps were characterized by
a high frequency of Primary support features including reduction, tunnelling, tailoring,
personalization and self-monitoring. Simulation and rehearsals were however absent
within this cluster. Dialogue support features such as reminders and suggestions had
the highest frequencies compared to praise, rewards and social role. Trustworthiness
was the only feature within Credibility support with the highest frequency, followed by
expertise and surface credibility. Real-world feel had the lowest frequency. Authority,
third party endorsement and verifiability were absent within this cluster. For Social
support features, social learning was the only feature present, and it had a high fre-
quency. In terms of opinion sentiments, this cluster also had a higher positive sentiment
value compared to negative sentiment value.
Ninth International Workshop on Behavior Change Support Systems (BCSS 2021): 33
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
Table 2 Classification of Emotional Words used for the emotional sentiments
Synonyms & relatable words
Liking affection, adoration, fondness, fond, liking, like, attraction, attracted, attract, caring, care, tenderness, tender, compassion, sentimentality, sentiment, lust, sexual, desire, passion, infatuation
lovely, excellent, good, loved, adore, best, perfect, magnificence, magnificent, yummy, love, wonderful, cheerfulness, cheerful, cheer, amusement, amuse, amused, amusing, bliss, blissful,
jolly, joviality, delight, enjoyment, glad, gladness, happiness, happy, jubilation, elation, elating, satisfaction, satisfying, satisfied, ecstasy, euphoria, zest, enthusiasm, enthusiastic, zeal, excite
exhilaration, exhilarating, content, contentment, pleasure, pride, triumph, optimism, optimistic, eager, eagerness, hope, enthrallment, rapture, relief, beautiful, enjoy, super, fantastic, superb, f
Trust trust, comfort, comfortable, encourage, encouraging, marvellous, marvel, kudos, thankful, perfect, perfection, friendly, friend, friendliest, ideal, flawless
Anger Irritability, irritating, irritated, irritation, aggravation, agitation, agitated, annoyed, annoyance, annoying, grumpy, crosspatch, exasperation, frustration, rage,
raging, annoy, anger, outrage, outraged, fury, wrath, hostility, hostile, ferocity, bitter, bitterness, hate, hatred, scorn, spite, vengeful, dislike, disliking, disliked,
resent, resenting, resentment, disgusting, disgust, revulsion, contempt, loathing, loathe, envy, jealous, jealousy, torment, tormenting, idiot, suck, sucker, loser,
Sadness Sad, suffer, suffering, pain, painful
Fear Horror, alarm, alarming, shocking, shock, fear, fearful, terror, panic, hysteria, mortification, nervous, nervousness, anxious, anxiety, suspense, uneasy, uneasiness,
suspenseful, apprehension, apprehend, worry, worrying, distress, distressful, dread, dreadful, danger, dangerous, fraud, hell, scam, stress, stressing, stressful
Table 3: The Intensity of Emotional and Opinion Sentiments
Features/App ID 2 3 5 6 8 9 10 18 19 22 23 4 11 15 16 20 17 21 1 7 12 13 14
Reduction 21
Primary Task
Tunnelling 20
Support
Tailoring 21
Personalization 21
Self-monitoring 19
Simulation 4
Rehearsal 3
Di-
Praise 11
logue
Sup-
port
a-
Rewards 10
34 Ninth International Workshop on Behavior Change Support Systems (BCSS 2021):
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
Features/App ID 2 3 5 6 8 9 10 18 19 22 23 4 11 15 16 20 17 21 1 7 12 13 14
Reminders 18
Suggestion 19
Social role 11
Trustworthiness 21
System Credibility
Expertise 15
Surface credibility 21
Support
Real-world feel 11
Authority 3
3rd party endorsement 5
Verifiability 10
Social learning 13
Social Support
Social comparison 5
Normative influence 8
Social facilitation 9
Cooperation 5
Competition 2
Recognition 3
Liking (%) 19.6 22.9 14.1 21.3 20.6 18.2 23.0 23.9 24.6 17.5 21.6 32.8 50.0 29.2 56.3 42.4 36.5 43.6 30.6 33.7 20.0 26.5 25.0
Sentiments
Trust (%) 5.0 5.7 2.9 5.0 6.7 5.3 6.7 8.0 8.5 4.4 5.4 7.5 10.7 8.3 3.1 3.0 6.4 2.6 6.7 8.4 4.8 6.8 9.8
Anger (%) 10.5 9.9 9.1 8.9 7.2 7.1 7.3 11.4 10.2 8.4 7.0 9.0 7.1 8.3 3.1 3.0 4.8 2.6 9.3 6.3 8.3 4.8 5.4
Sadness (%) 4.1 3.1 5.6 4.0 5.0 3.1 2.8 6.8 4.2 4.7 3.2 0.0 0.0 1.0 0.0 0.0 3.2 2.6 4.0 2.1 3.9 5.4 4.4
Fear (%) 2.7 3.7 3.7 3.5 2.2 4.0 2.8 1.1 1.7 4.0 3.8 6.0 3.6 5.2 3.1 6.1 0.0 0.0 2.7 2.1 3.5 2.0 4.4
Positive (%) 61.8 66.7 46.7 60.4 65.6 64.4 65.7 60.2 66.9 59.6 62.7 64.2 85.7 75.0 90.6 72.7 79.4 84.6 65.3 72.6 60.4 72.1 64.1
Ninth International Workshop on Behavior Change Support Systems (BCSS 2021): 35
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
Features/App ID 2 3 5 6 8 9 10 18 19 22 23 4 11 15 16 20 17 21 1 7 12 13 14
Negative (%) 38.2 33.3 53.3 39.6 34.4 35.6 34.3 39.8 33.1 40.4 37.3 35.8 14.3 25.0 9.4 27.3 20.6 15.4 34.7 27.4 39.6 27.9 35.9
C1 C2 C3 C4 C5 C6
36 Ninth International Workshop on Behavior Change Support Systems (BCSS 2021):
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
Ideal Weight (app1), Cycling – bike tracker (app7), Water Drink Reminder (app12),
Step Counter – Calorie Counter (app13) and Weight Loss Running (app14) were found
to have higher frequencies of tailoring, personalization and self-monitoring as primary
support features. In this cluster (i.e., C6) however, simulation had the lowest frequency
and rehearsal was absent. Dialogue support features including rewards, reminders and
suggestions were marginally present with reward having the lowest frequency. Also,
praise and social role features were absent. With regard to Credibility support, exper-
tise, real-world feel, authority and verifiability were absent while trustworthiness and
surface credibility were present with high frequencies. See table 3 for details of the
various clusters and their corresponding sentiments (clusters are differentiated with dif-
ferent fills and patterns).
4.3 Implication of Study
Generally, the findings revealed that health and fitness apps are popular since user re-
views are mostly positive. Almost all the apps had high positive sentiments. Some ap-
plications recorded positive sentiments above 90% and this is promising for HBCSS
research and practice. With regard to emotional sentiments (i.e., Liking, Trust, Anger,
Sadness and Fear), the findings revealed that most users expressed some form of like-
ness for the apps. Sentiment words that exhibit likeness were observed in most of the
reviews. However, an analysis of the various clusters of apps in relation to system fea-
tures showed that the provision of more persuasive features does not guarantee favour-
able sentiments from users. This is because, apps that had more system features did not
record higher emotional sentiments. For instance, Pocket Yoga (app16) had only one
persuasive feature (i.e., reduction), yet it recorded the highest emotional sentiment in-
tensity. Also, it had the highest positive sentiment intensity. It recorded lower ratings
for Trust, Anger, Fear, and Sadness. This demonstrates that although the absence of
Credibility support features leads to a lack of trust, credibility support has less impact
on application acceptance (likeness). Also, the presence of more features does not guar-
antee specific sentiments (i.e., no clear pattern between system features and senti-
ments). It can be argued that the presence of more persuasive features rather provides
users with the opportunity to assess each functionality as compared to fewer features.
Hence, applications with more persuasive features appear complex to users and there-
fore do not attract high sentiments of likeness.
The study also revealed that the presence or absence of Credibility support features
does not guarantee trust in user sentiments. Considering that Credibility support fea-
tures seek to promote system trust, this finding is worrying. It was observed that apps
including WalkingApp (app2), Headspace (app5), and HidrateSpark Smart Bottle
(app18) had relatively high Credibility support features, yet they recorded lower trust
sentiments when compared to Cycling - Bike Tracker (app7) and Weight Loss Running
by Verv (app14). It was also observed that applications that had more Social support
features had more sentiment words that demonstrate anger, fear and sadness when com-
pared to those with no or less social support features. For instance, apps such as Walk-
ingApp (app2), Headspace (app5), HidrateSpark Smart Bottle (app18), and Six Pack
in 30 Days (app22) had more social support features present and they also recorded
Ninth International Workshop on Behavior Change Support Systems (BCSS 2021): 37
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
higher sentiment words when compared to Cycling - Bike Tracker (app7), Water Drink
Reminder (app12), Step Counter - Calorie Counter (app13), and Weight Loss Running
by Verv (app14) that had less Social support features.
5 Conclusion
This study presents findings from an investigation of the relationship between user sen-
timents and persuasive system features. It adopted a stratified random sampling tech-
nique to select health and fitness apps on the Android and iOS markets. The sentiments
of app users were extracted and compared with systems features that are available in
each app using clustering techniques. The results demonstrated that the provision of
more persuasive features does not guarantee favourable sentiments from users. Partic-
ularly, it was observed that apps with less system features attracted more sentiments
relating to likeness. Also, it was observed that Social support features mostly promote
negative emotions such as anger, fear and sadness. Perhaps, these findings corroborate
with existing knowledge that argues that the presence of persuasive system features in
Health Behaviour Change Support Systems (HBCSSs) do not necessitate the suffi-
ciency or efficiency of the system [13]. More importantly, there is a need for further
investigation to be conducted to explain the causal effects of this phenomenon. Partic-
ular attention must be given to the type and structure of messages used in conveying
the various persuasive features. This is because although designers of persuasive appli-
cations may convey the intention to change in their messages, the messages may gen-
erate an emotional shift from their intentions.
References
1. H. Oinas-Kukkonen and M. Harjumaa, “Persuasive systems design: Key is-
sues, process model, and system features,” Communications of Association of
Information Systems, vol. 24, no. 1, p. 28, 2009.
2. I. Wiafe, F. N. Koranteng, F. A. Kastriku, and G. O. Gyamera, “Assessing the
impact of persuasive features on user’s intention to continuous use: the case
of academic social networking sites,” Behaviour and Information Technology,
vol. 0, no. 0, pp. 1–19, 2020, doi: 10.1080/0144929X.2020.1832146.
3. J. Dabi, I. Wiafe, A. Stibe, and J. D. Abdulai, “Can an enterprise system per-
suade? The role of perceived effectiveness and social influence,” in Lecture
Notes in Computer Science (including subseries Lecture Notes in Artificial
Intelligence and Lecture Notes in Bioinformatics), vol. 10809 LNCS, 2018,
pp. 45–55.
4. T. Lehto and H. Oinas-Kukkonen, “Explaining and predicting perceived ef-
fectiveness and use continuance intention of a behaviour change support sys-
tem for weight loss,” Behaviour and Information Technology, vol. 34, no. 2,
pp. 176–189, 2015, doi: 10.1080/0144929X.2013.866162.
5. I. Wiafe, M. M. Alhammad, K. Nakata, and S. R. Gulliver, “Analyzing the
persuasion context of the persuasive systems design model with the 3D-RAB
38 Ninth International Workshop on Behavior Change Support Systems (BCSS 2021):
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
model,” in Lecture Notes in Computer Science (including subseries Lecture
Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2012,
vol. 7284 LNCS, pp. 193–202, doi: 10.1007/978-3-642-31037-9_17.
6. I. Wiafe, “The Role of U-FADE in Selecting Persuasive System Features,”
Encyclopedia of Information Science and Technology (4th ed.). pp. 7785–
7795, 2017, doi: 10.4018/978-1-5225-2255-3.ch677.
7. I. Adaji and J. Vassileva, “Perceived effectiveness, credibility and continuance
intention in e-commerce: a study of Amazon,” in International Conference on
Persuasive Technology, 2017, pp. 293–306.
8. K. Oyibo, R. Orji, and J. Vassileva, “Investigation of the social predictors of
competitive behavior and the moderating effect of culture,” UMAP 2017 - Ad-
jun. Publ. 25th Conf. User Model. Adapt. Pers., no. October, pp. 419–424,
2017, doi: 10.1145/3099023.3099113.
9. D. DeSteno, D. T. Wegener, R. E. Petty, D. D. Rucker, and J. Braverman,
“Discrete Emotions and Persuasion: The Role of Emotion-Induced Expectan-
cies,” J. Pers. Soc. Psychol., vol. 86, no. 1, pp. 43–56, 2004, doi:
10.1037/0022-3514.86.1.43.
10. B. J. Fogg, “A behavior model for persuasive design,” Proc. 4th Int. Conf.
Persuas. Technol. - Persuas. ’09, p. 1, 2009, doi: 10.1145/1541948.1541999.
11. S. Langrial, T. Lehto, H. Oinas-Kukkonen, M. Harjumaa, and P. Karppinen,
“Native mobile applications for personal wellbeing: A persuasive systems de-
sign evaluation,” in Proceedings - Pacific Asia Conference on Information
Systems, PACIS 2012, 2012, no. January.
12. I. Wiafe and K. Nakata, “A Semiotic Analysis of Persuasive Technology: An
Application to Obesity Management,” 2012.
13. T. Lehto and H. Oinas-Kukkonen, “Examining the Persuasive Potential of
Web-based Health Behavior Change Support Systems,” AIS Trans. Human-
Computer Interaction., vol. 7, no. 3, pp. 126–140, 2015, doi:
10.17705/1thci.00069.
14. T. Lehto, H. Oinas-Kukkonen, and F. Drozd, “Factors Affecting Perceived
Persuasiveness of a Behavior Change Support System,” in Proceedings of the
International Conference on Information Systems (ICIS), 2012, pp. 1–15.
15. R. E. Petty and J. T. Cacioppo, “The elaboration likelihood model of persua-
sion,” Adv. Exp. Soc. psychology, vol. 19, pp. 123–205, 1986.
16. G. J. Stigler, “The economics of information,” J. Polit. Econ., vol. 69, no. 3,
pp. 213–225, 1961, doi: 10.1300/J111v26n01_07.
17. P. Ekman, “Basic emotions. Handbook of cognition and emotion,” in Hand-
book of Cognition and Emotion, New York, New York, USA: John Wiley &
Sons, Ltd, 1999, pp. 301–320.
18. D. Keltner and J. S. Lerner, “Emotion.,” in Handbook of social psychology,
Vol. 1, 5th ed., Hoboken, NJ, US: John Wiley & Sons, Inc., 2010, pp. 317–
352.
19. J. So, C. Achar, D. H. Han, N. Agrawal, A. Duhachek, and D. Maheswaran,
“The psychology of appraisal: Specific emotions and decision-making,” J.
Consum. Psychol., vol. 25, no. 3, pp. 359–371, 2015, doi:
10.1016/j.jcps.2015.04.003.
20. W. N. W. Ahmad and N. M. Ali, “Engendering trust through emotion in
Ninth International Workshop on Behavior Change Support Systems (BCSS 2021): 39
Exploring the Impact of Persuasive System Features on User Sentiments in Health and Fitness Apps
designing persuasive application,” in Lecture Notes in Computer Science (in-
cluding subseries Lecture Notes in Artificial Intelligence and Lecture Notes in
Bioinformatics), vol. 8237 LNCS, 2013, pp. 707–717.
21. J. P. Dillard and R. L. Nabi, “The persuasive influence of emotion in cancer
prevention and detection messages,” J. Commun., vol. 56, no. 2006, pp. 123–
139, 2006, doi: 10.1111/j.1460-2466.2006.00286.x.
22. R. L. Nabi, “Emotional Flow in Persuasive Health Messages,” Health Com-
mun., vol. 30, no. 2, pp. 114–124, 2015, doi: 10.1080/10410236.2014.974129.
23. W. N. W. Ahmad and N. M. Ali, “Trust perceptions in using persuasive tech-
nologies,” in 3rd International Conference on Computer and Information Sci-
ences, ICCOINS 2016 - Proceedings, 2016, pp. 49–53, doi:
10.1109/ICCOINS.2016.7783187.
24. K. Halttu and H. Oinas-Kukkonen, “Persuading to Reflect: Role of Reflection
and Insight in Persuasive Systems Design for Physical Health,” Human-Com-
puter Interact., vol. 32, no. 5–6, pp. 381–412, 2017, doi:
10.1080/07370024.2017.1283227.
25. W. N. W. Ahmad and N. M. Ali, “A Study on Persuasive Technologies: The
Relationship between User Emotions, Trust and Persuasion,” Int. J. Interact.
Multimed. Artif. Intell., vol. 5, no. 1, pp. 57–61, 2018, doi:
10.9781/ijimai.2018.02.010.
26. P. Indraratna et al., “Mobile phone technologies in the management of is-
chemic heart disease, heart failure, and hypertension: systematic review and
meta-analysis,” JMIR mHealth uHealth, vol. 8, no. 7, 2020, doi:
10.2196/16695.
27. K. T. Win, M. R. H. Roberts, and H. Oinas-Kukkonen, “Persuasive system
features in computer-mediated lifestyle modification interventions for physi-
cal activity,” Informatics Heal. Soc. Care, vol. 44, no. 4, pp. 376–404, 2019,
doi: 10.1080/17538157.2018.1511565.
28. M. M. Alhammad and S. R. Gulliver, “Persuasive technology and users’ ac-
ceptance of e-commerce: users’ perceptions of website persuasiveness,” J.
Electron. Commer. Organ., vol. 12, no. 2, pp. 1–13, 2014.