- Email: [email protected]
Assessing mobile health applications with twitter analytics
Accepted Manuscript Title: Assessing mobile health applications with twitter analytics Authors: Rajesh R. Pai, Sreejith Alathur PII: DOI: Reference:
S1386-5056(18)30119-9 https://doi.org/10.1016/j.ijmedinf.2018.02.016 IJB 3666
To appear in:
International Journal of Medical Informatics
Received date: Revised date: Accepted date:
13-9-2017 20-2-2018 22-2-2018
Please cite this article as: Rajesh R.Pai, Sreejith Alathur, Assessing mobile health applications with twitter analytics, International Journal of Medical Informatics https://doi.org/10.1016/j.ijmedinf.2018.02.016 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ASSESSING MOBILE HEALTH APPLICATIONS WITH TWITTER ANALYTICS
Author names and affiliations: Rajesh R. Paia,*, Sreejith Alathurb a,b
School of Management, National Institute of Technology Karnataka, Surathkal,
IP T
India
SC R
[email protected]
*Corresponding author:
U
Rajesh R. Pai
N
School of Management, National Institute of Technology Karnataka, Surathkal,
A
India
M
[email protected]
TE D
Highlights
Sentiment analysis was performed on the Twitter data using the keywords viz. fitness apps, diabetes app, meditation app, and cancer app representing the types of mobile health applications.
Perception and usage experience of mobile health applications among the citizens has been
EP
studied and identified that except for cancer application there exists a positive polarity
CC
towards the fitness, diabetes and meditation application among the users.
System thinking approach followed with the help of causal loop diagram for identifying
A
the relationships between the variables influencing the accessibility and acceptability of mobile health application among the citizens, healthcare, patients and application developers which help them in understanding the dynamics involved in system.
Introduction: Advancement in the field of information technology and rise in the use of Internet has changed the lives of people by enabling various services online. In recent times, healthcare sector which faces its service delivery challenges started promoting and using mobile health applications with the intention of cutting down the cost making it accessible and affordable to the
IP T
people.
Objectives: The objective of the study is to perform sentiment analysis using the Twitter data
SC R
which measures the perception and use of various mobile health applications among the citizens.
Methods: The methodology followed in this research is qualitative with the data extracted from a social networking site “Twitter” through a tool RStudio. This tool with the help of Twitter
U
Application Programming Interface requested one thousand tweets each for four different phrases
N
of mobile health applications (apps) such as “fitness app”, “diabetes app”, “meditation app”, and
M
emotions were measured.
A
“cancer app”. Depending on the tweets, sentiment analysis was carried out, and its polarity and
TE D
Results: Except for cancer app there exists a positive polarity towards the fitness, diabetes, and meditation apps among the users. Following a system thinking approach for our results, this paper also explains the causal relationships between the accessibility and acceptability of mobile health
EP
applications which helps the healthcare facility and the application developers in understanding and analyzing the dynamics involved the adopting a new system or modifying an existing one.
CC
Keywords: Mobile Health, Sentiment Analysis, Twitter Analytics, Causal Loop Diagram, Technology Adoption Model
A
1. Introduction Health is considered to be the most important component in the lives of every human being. Given such a priority, maintaining good health with healthy diet, moderate exercises, and regular medical
checkups are the greatest challenges which people are facing currently in this busy world. Moreover, with the large geographical area and population level, the dichotomy in demand for the healthcare and lack of health insurance coverage resulted in increased cost of the healthcare services. Consequently, this has created an impact on providing better healthcare and treatment
IP T
services across rural and urban cities and eventually has become a major concern for many
governments worldwide [1]. To reduce the disparity of population level (5,300 million) and
SC R
healthcare (11 million), various governments and healthcare sectors, are promoting awareness about mobile technology use in healthcare services for the people living in the rural and urban
regions [2]. The use of mobile technology in communicating healthcare practices to the citizens
U
through health applications (apps) are often termed as mobile health (mHealth). It was first defined
N
by Robert Istepanian as “emerging mobile communications and network technologies for
A
healthcare systems” [3]. This technology is considered to be the backbone of electronic health
M
(eHealth) due to its ability to provide healthcare access to the people living in rural areas. These
TE D
apps act as interphase between the people and the physician/doctors for consultations and monitoring through videos or messages facilitating them in maintaining health. Mobile health has been emerged due to rapid growth in the use of mobile technologies and its
EP
connectivity to access health-related information. Thus transforming the present health service delivery practices to more of non-physical interaction between Healthcare Providers (HCP) and
CC
patient/user [4]. This is required especially for the people living in a low-resource setting wherein access to healthcare facilities and services are often minimal. To tackle this, government and
A
technology entrepreneurs are promoting various mHealth apps with the intention of reducing healthcare service cost making it convenient for emergencies and general health services. Even though some people are using mHealth apps on a regular basis for maintaining their health,
the data results or evidence regarding health outcomes are still not clear. In such cases, microblogging sites due to the nature and accessibility of data availability can be used to collect peoples’ opinions about these apps. These data along with consumer health informatics can be used for generating real-time information from user posts relating to the products, service,
IP T
individual, events, and place which helps the user in building trust towards the use of mHealth
apps. Currently, there are various microblogging sites such as Twitter, Google+, Facebook, etc.
SC R
however, Twitter and the Facebook are the most commonly used.
Therefore, the current study uses Twitter data for analyzing the people’s sentiment, as it is considered to be the most popular page rank microblogging sites of the year 2016 depending on
U
the number of users and page views [5]. A user here on this site can post a message or tweet relating
N
to the product or an event with a word limit of 140 characters. Moreover, it can also be used as an
A
online news service portal through which individuals can view discussions which are happening
M
across the globe.
TE D
Many times, these tweets play an important role in manufacturing and service organizations to study the user's reaction or responses about its products or services which helps them in making decisions. However, the greatest challenge is to organize these data so as to transform it into useful
EP
facts and draw conclusions [6]. Thus the analysis performed in this study extracts tweets and analyzes user’s sentiments or emotions by classifying them into positive, negative or neutral using
CC
a tool called as RStudio through different R-codes. The current study makes an attempt to access the sentiment score for different mHealth apps
A
through a social networking site called Twitter, and thereupon causal relationships have been established to identify the interrelationships between the factors.
2. Literature Review Mobile technologies are growing dramatically in recent years due to the advancement in the field of information and communication technologies. This has changed the lifestyle of the people through different mobile applications (apps) which in turn changed the way of communication and
IP T
living. Since then, mobile technologies are in continuous use for delivering a large range of
services starting from calling, messaging, etc. up to its utility in banking sectors due to the features
SC R
such as speed of accessibility and /or mobility. Currently, it has entered into healthcare sectors
with different features coupled with computer-supported technologies and medicine. In healthcare,
doctors for monitoring patient and delivering care [7].
U
these apps find its use in collecting community health data and its delivery to practitioners, and
N
According to the study conducted by various research associations, there are more than 165,000
A
mobile health (mHealth) apps available across different mobile app stores [8]. Most of these apps
M
finds its use in maintaining Health and Fitness (such as LoseIt, Google fitness, Fitbit, etc.), Diet
TE D
Companion Apps (Health and Nutrition, Calorie counter etc.), Stress Relief Apps (Simple Yoga, Relax meditation: Sleep sounds, etc.), Diabetes apps (mySugr Diabetes Logbook, Glucose Buddy: Diabetes Log, etc.) etc. Concerning patient health record, it was identified that doctors and patients
EP
have different attitudes and expectations which found to have significant sociodemographic, educational and attitude correlation with patient-held health records [9]. Additionally, electronic
CC
and other Information Technology (IT) system adoption in a complex healthcare environment slows- down process as it doesn’t have one size fits all devices and requires consideration of many
A
healthcare and technology dimension [10]. Jiang et al. [11], studied the response of lung transplant recipients towards a mHealth app ‘Pocket PATH®’ which was designed to provide automatic feedback messages detecting and reporting
critical values of health and found to be appropriate among the recipients. Likewise, when the app is used for hypertensive and diabetic patients, a phone call or SMS reminders was an acceptable and favorable option for managing the chronic diseases [12]. However, research is still in process with various techno-entrepreneurs and academicians granting maximum access and support to the
IP T
people in enhancing their health behaviors.
Many researchers have explained the potential of using mHealth in supporting health workers and
SC R
citizens of the low and medium income settings. However, the evidence base with respect to
documentation, designs, and reported outcomes are weak and inconsistent. Hence, its utility can be measured by studying the complexity of the system involved in the mHealth adoption by
U
understanding the nature of sentiments people carry during its purchase and use. Thus, this section
A
(called sentiment analysis) for the Twitter data.
N
discusses the use of system dynamics in the mHealth adoption followed by social media analytics
M
2.1. System Dynamics and Mobile Health Adoption
TE D
Sentiment score executed from the Twitter sentiment analysis can be used with the set of differential equation to study the interactions between the dependent and independent variables of a system which is defined. In addition, system thinking approach can also be followed in the form
EP
of causal loop diagram (CLD) so as to measure the factors influencing mHealth adoption and accessibility among the people. These CLD’s are represented with different feedback structures of
CC
the systems by capturing individuals or teams mental model or communicating the feedbacks which are highly responsive causing dynamics [13]. It can also be used in standard system dynamic
A
practices for connecting the simulation analysis and modeling to study the behavior of the system over time [14]. System Dynamics mathematics with tweets sentiment values can be used to predict smartphone
sales helping the company to analyze customer perception about their offering [15]. Sedarti and Batash [16], used CLD to understand human behavior involved in the dynamics of wearable smart devices use and identified social influence as a significant factor for the device acceptance. The dynamics of the user behavior was also studied by extending the hypothesis and incorporating
IP T
various feedbacks leading from Information System (IS) use to perceived ease of use on one side
and productivity to IS-related work on the other. These feedbacks help managers in devising
SC R
strategies during system changes or introduction of new systems [17].
Consequently, the managers require in-depth knowledge about technology adoption and their behavior when influenced by adopters and inhibitors depending upon design and technology
U
evaluation policies. Therefore concluded that peak adoption time delay and peak adoption rate
N
could be affected by the adoption inhibitors and the diffusion process of technology [18]. This is
A
because of potential and current technology adopters who are unaware of the benefits of
M
technology adoption which they learn through information feedback present in the system. In such
TE D
cases, system dynamics modeling can be used to overcome facilitating factors of time and adoption [19].
Chen [20], identified that organization would accept technology based on management attitude
EP
and perception which can be made favorable through effective promotions and word-of-mouth. Likewise, customers also consider the return on investment, time and effort from that technology.
CC
Wong and Sheng [21] identified that word-of-mouth with moderating factors such as word-ofmouth valence, trust, and risk attitude reduces uncertainty involved in purchasing or consuming
A
any products and services. Liu et al. [22], investigated the adoption of mobile government among the rural population and identified that young males having knowledge about policies have positive perception and become
potential adopters of the technology. They can later be transformed into adopters through wordof-mouth, perceived ease of use, usefulness, image and social influence. Moreover, technology adoption is also dependent on the factors such as subjective norm, health consciousness, health information orientation, electronic health literacy, internet health information use efficacy,
IP T
perceived usefulness, perceived ease of use, and behavioral intention to use when young student populations are considered [23].
SC R
2.2. Social Media Analytics and Mobile Health
Social media analytics came into the visibility due to availability and accessibility of a large amount of data which organizations across the world are finding it difficult to manage and derive
U
a meaningful solution. Therefore, various analytics has been identified among them sentiment
N
analysis is the one which assists organizations in establishing data relationships.
A
Sentiment analysis or opinion mining focuses on the computational study of opinions, sentiment,
M
and emotions which are expressed in the form of text [24]. In other words it refers to the processing
TE D
of natural language, text analysis, and biometrics to identify, extract, quantify, study the affective states, and subjective information about a product/service, or an event from the perspective of the customer through reviews, survey responses obtained from the social media or blogs [25]. These
EP
responses are helpful for the organizations to measure the sentiment of an individual having towards their products or services which assists them in production and marketing.
CC
In psychology, there exist different theories which explain peoples’ need and emotions towards the mHealth app use for health activities. But then most of these depend on how we perceive our
A
social environment [26] and this may not be constant until people get experiences over time. Hence, the following section reviews select literature which drives peoples’ needs and emotions for the use of mHealth apps.
Theories of needs and motivation: According to Abraham H. Maslow, people will be motivated to use products only when their needs are unmet or unsatisfied [27]. These unsatisfied need depends upon satisfaction which influences the behavior driving them to progress upwards in the hierarchy level as classified by Alderfer’s in
IP T
ERG theory of Existence (need physiological health and safety), Relatedness (need for
interpersonal connections, social status, and recognition) and Growth (need for personal
SC R
development) [28]. These motivations can also arise because of needs associated with achievement, power, and affiliation as David McClelland explained in his three needs theory [29].
Therefore, in the case of mHealth app use the primary reason being the disparity in the level of
U
population and healthcare accessibility resulted in problems related to the delivery of good quality
N
healthcare services. Since these needs are unmet, research associations, healthcare units, and the
TE D
Theories of emotions:
M
information’s over mobile phones.
A
government started promoting health apps such that a number of people can have access to health
Emotions are the feeling which a person expresses depending on circumstances or environment. It can be any conscious experiences characterized by intense mental actions resulting in high level
EP
of joy or disappointment [30]. In evolutionary theories, emotions such as fear, jealousy, anxiety, anger, desire, love, etc. had evolved from the early hominids, using certain adaptations such as
CC
incorporation, rejection, destruction, protection, reproduction, reintegration, orientation, and exploration, which are being shared by animals, resulting in the processing of certain traits due to
A
their shared ancestry [26, 31-34]. Social and cultural theories of emotions consider emotions which starts with an idea and a product of societies and culture acquired by the individuals through experience [26]. This theory also
addresses emotion words used in different languages for representing a particular behavior or situation [35, 36]. Example: in Japanese, the word ‘Fago’ describes maturity level of the person like childishness, selfishness, etc. and ‘Song’ refers to unacceptable behavior [37]. Further, emotion process starts from the perception of stimulus then triggers a response of the body and
IP T
finally body responses in terms of change in blood pressure, heartbeat, facial expression, etc. i.e.,
conceptualization of anxiety disorders [26, 38]. For example, anger because of insult and joy
SC R
because of good grades in exams.
In judgment theories of emotion process, emotions such as surprise, hope, fear, joy, relief, discomfort, disgust, etc. are considered to have a basic judgment about ourselves and our place in
U
the world based on a projection of ethics and values, structures and mythologies. However, the
N
cognitive theories signify the way the individuals evaluate the situation determines the emotions
A
[39, 40].
James-Lange Theory: It proposes that the people will experience the emotions when they
TE D
M
Some of these emotion theories as defined and distinguished by David G. Myers [41] were:
perceive things. Example: When we feel sad, we cry, likewise when we feel embarrassed we blush.
Cannon-Bard Theory: This theory suggests that people will experience emotions and
EP
physiological arousal at the same time, which means that emotions and arousal are
Schachter-Singer “Two-factor” Theory: It says people will experience emotions which depend
A
CC
simultaneous. Example: We blush and feel embarrassed at the same time simultaneously.
upon physiological arousal and its cognitive interpretation. Example: The label which people give for the same physiological arousal at different instances and context like fear can become arousal depending on the environment.
Robert Zajonc, Joseph LeDoux, and Richard Lazarus: Emotions without Awareness/Cognition Theory: This theory states that some reactions which are emotional develop on a low road through brain skipping the thought of consciousness. Example: In the forest, we respond to the sound before assessing it.
IP T
Based on these theories many researchers across the world have studied the sentiments or emotions of the people they followed while writing books, blogs, social networking sites, etc. Among them,
SC R
some of their findings have been explained in the literature below.
Pagolu et al. [42], found that there is a strong correlation between the stock price and public
U
opinions by analyzing Twitter data. The Twitter data is also analyzed based on volume, generic sentiment, and language model for predicting the movement of the public opinion measured
N
through presidential approval rating, economic confidence, and the generic Congressional ballot
A
[25]. Lassen et al. [15], used sentiment analysis by developing a linear regression model for
M
predicting the sales of the iPhone from its tweets which are found to have a strong correlation.
TE D
Twitter can also be used to analyze the sentiment of the people about particular healthcare or public health measures. Rastegar-Mojarad et al. [43], used sentiment analysis techniques and Hadoop to analyze the patient experience of healthcare and concluded that longer the patient reviews poorer
EP
the reviews of the facility. Twitter analysis can also be used to measure the real-time barometer of public on Affordable Care Act (ACA) and identified that there exists a positive correlation between
CC
the sentiment and marketplace enrollment [44]. Paul and Dredze [45], correlated influenza rates
A
with the Twitter data and indicated that for a particular symptom of infections or physical injuries the intended usage of medicine can be mentioned. It can also be used for extracting and investigating the effect of adverse drug reactions helping them in improving drug and creating popularity in the social media and other health forums [46]. Salathe´ and Khandelwal [47],
investigated the spatiotemporal sentiments for a novel vaccine which can be used to measure the intuition about that vaccine. Hence they concluded that groups of negative vaccine sentiments leads to unprotected individual and increases disease outbreaks. Considering these literature, this research paper uses Twitter data to study the people’s perceptions and experiences towards the
IP T
mHealth app use. 3. Methodology
SC R
The methodology followed in this research is qualitative with the data extracted from a social
networking site “Twitter” using RStudio. The framework for the tweets extraction is represented
U
in figure 1. 3.1. Dataset
N
RStudio tool is used to extract Twitter access using ROAuth and web API (Application
A
Programming Interface) and requested 1000 tweets each for four different phrases of mHealth
M
apps, namely fitness app, diabetes app, meditation app, and cancer app. Tweets were extracted for
TE D
latest 8 days starting from 08 June 2017 to 15 June 2017, and sentiment analysis was performed to measure its polarity and emotions. 3.2. Sentiment analysis of tweets
EP
This paper uses word cloud, and bigram for variable frequency identification found on the Twitter website. From this platform, using R codes tweets with keywords viz. fitness app, diabetes app,
CC
meditation app, and cancer app were extracted, and sentiment analysis was performed. Table 1,
A
indicates tweets requested and obtained for each of the mHealth apps: 3.3. Causal loop diagram for mHealth adoption Causal Loop Diagrams (CLD) visualizes the cause and effect relationships between variables involved in mHealth devices adoption among citizens. CLD’s are represented by polarities
(positive (+) and negative (-)) and delay. A positive polarity between ‘A’ and ‘B’ indicates as ‘A’ increases ‘B’ also increases and a negative polarity represents as variable ‘A’ increases, variable ‘B’ decreases in the opposite direction [13]. It is used in this study considering the variables of Technology Adoption Model (TAM) of Davis et al. [48]. This model considers perceived
for measuring the intention to use and actual system use or adopters.
SC R
4. Result and Analysis
IP T
usefulness, perceived ease of use, and attitude of using the technology as the major components
The RStudio requested 1000 tweets for which bigram, word cloud, and sentiment scores were
U
calculated under four different cases depending upon the type of mHealth apps. Case i: Twitter analytics for fitness applications
N
Figure 2a and 2b indicate that people using iPhones are aware of fitness apps which are new and
A
freely available and are discussing its importance in managing services at a clinic or for
M
maintaining a balanced diet (food diary, nutrition, etc.). Moreover, users may also consider if
TE D
suggested by clinics or hospitals such as Mayo clinic, etc. Figure 3a specifies that about 74% of the tweets have positive sentiments, indicating that people are seriously considering fitness apps as a tool for improving their health. It is supported by figure
EP
3b, indicating application users’ emotion as joyful (132 tweets) and higher when compared other emotion components.
CC
The following Table 2 below shows select tweets with its polarity and emotion values.
A
Case ii: Twitter analytics for diabetes applications Figure 4a and 4b indicate that people may be aware of diabetes mobile apps used for monitoring glucose level in the blood through one-touch digital health meters revealing the type of hypoglycemia (type 1 and type 2) among the people.
The total number of tweets for diabetes app extracted during the analysis period was only 365 numbers. This value appears to be less than fitness application and may improve over time as the positive sentiment tweets are 259 representing 26% of the total tweets extracted (figure 5a). The reason behind this may be because of awareness or interoperability issues between users and
IP T
doctors in establishing interphase resulted in sadness. Alternatively, people using this app frequently on a daily basis are found to be joyful in using the application (figure 5b).
SC R
The following Table 3 provides the analysis of tweets extracted based on emotion and polarity. Case iii: Twitter analytics for meditation applications
U
Figure 6a and 6b illustrate that people often talk about these applications due to its mindfulness
N
and deep sounds features which are binaural and available freely in mobile play stores. This
A
application helps the user to feel relaxed through their audios and improve concentration helping
M
them in brain development.
From the figure 7a, we can identify that there is a positive opinion about the meditation app among
TE D
people as the total number of positive sentiment is 810, which corresponds to 81% of the total tweets extracted. Results of the sentiment analysis also indicate that 8.1% of the people/user are joyful in their perception and use of this app (figure 7b). This signifies that meditation app is
EP
working fine and people are satisfied with it. Some of the select comments about this applications
CC
are shown in the table below. Case iv: Twitter analytics for cancer applications
A
Figure 8a and 8b show that people are discussing about cancer application available in different play stores but are much concerned about the survival benefits as it is paradoxical in nature with an example being ‘a cough turned out to be lung cancer via the app.' Many users suggest that use of mobile cancer application in the early stages of diagnosis by displaying some documented
symptoms can be helpful in improving and saving lives. For the cancer application, the number of positive sentiments is less compared to that of negative (figure 9a). This indicates that the perception and use of this application appear to be less as most of the people may prefer visiting the doctors personally than through smartphones for their regular
IP T
checkups. Though, there are more negative sentiment values people have expressed joyful (60
tweets) when compared with other emotional parameters (figure 9b). This result can be because of
SC R
features such as diet tracker, first aid process, finding the doctor and for booking appointments,
etc. which are available in this application. Table 5 below shows the examples of different tweets
U
extracted from various cancer applications.
N
Boxplot and sentiment scores for mobile health applications
A
The following figure (figure 10a and 10b) shows the boxplot indicating interquartile range values
M
and sentiment scores for four different mHealth applications.
TE D
Considering cancer app (figure 10a), we can identify the existence of positive skewness (median less than mean) in the negative region as the data are distributed below the median values of the boxplot. This indicates that people perception or use of cancer app is less than the average
EP
perception or use which is also identified in the figure 10b. Table 6 shows the boxplot for mHealth
A
CC
application indicating median, quartile and adjacent values.
Moreover, for diabetes app, the median is more than the mean (negative skewness) but in the positive region. These extreme positive sentiments are considered to be good for the company
because of the people’s positive perception towards the app use (figure 10a). This can also be identified from the figure 10b as more than 50% of the data falls onto the right side or positive side of the graph. Positive skewness is also seen even in the case of fitness and meditation app (figure 10a) with the
IP T
data in the positive region having almost the same sentiments value with progressive signs about
the apps as 50% of data falls on the positive side (figure 10b). This means that people realized the
SC R
importance and tended to use these applications to maintain their health which can, in turn, help the company to update their health applications to remain competitive.
Systems thinking approach for analyzing mobile health application adoption
N
feedback loops for understanding systems’ behavior.
U
From the figure 11 and figure 12, we can consider the presence of reinforcing and balancing
A
[Insert figure 11] (2-column fitting image)
M
Figure 11: Causal loop diagram showing the relationship between the potential adopters, semi-
TE D
adopters, and adopters of mHealth system
[Insert figure 12] (2-column fitting image)
EP
Figure 12: Causal loop diagram for perceived usefulness of mHealth system The first reinforcing loop R1, named as “benefits from the use of mHealth loop” connects the user
CC
and mHealth application willingness in increasing the completion rate. This means that more the number of mHealth users, more they satisfy with the application resulting in effective utilization
A
of mHealth system thus achieving desired health target. At the point of time when the user gets benefitted, their desire for mHealth also increases. This encourages healthcare providers or application developers to put efforts on the quality improvement of the information system which improves the overall system quality. When this increases, the user may feel that the system is
becoming more useful and easier to use which leads to dependent and willing to use. The second reinforcing loop R2, named as “users perception of quality and usage” basically explains, more the number of users more they will satisfy with the system and feels that the system is becoming more useful increasing the attitude of using health application more than they used
IP T
earlier.
The third reinforcement loop R3, named as the “learning loop” is developed based on user learning
SC R
which takes place when the user actually uses the system. In other words, more the number of users more the learning taking place and with electronic health (eHealth) literacy develops self-
efficacy or confidence in achieving targets easily thereby increasing attitude and willingness to
U
use the system.
N
“Market saturation loop (R4)” and “word of mouth loop (R5)” focuses on transforming new and
A
potential adopters to semi-adopters which later become the adopters of the system. The story
M
behind these loops represents, as the potential adopter of the system through proper promotional
TE D
take-ups (by creating awareness) improves the take-up rate which also increases along with the adopters contact through word of mouth take-up resulting from mHealth system/application adoption. This increase in the take-up rate reduces the number of potential adopters thereby
EP
reducing the adoption or use of the system. Moreover, the take-up rate is also influenced by the accessibility factors, i.e., IT infrastructure, technology literacy and socioeconomic dimensions
CC
(such as age, income, and education) relating to mHealth system or application. The reinforcement loops R6 and R7 represents the transformation of the users from semi-potential
A
adopters to adopters (or actual system use) through practice rate and learning on continuous use of the system. In a mHealth system, more the number of adopters or users more the ease in patient data collection, facilitating data process assisting healthcare providers in business dynamicity
thereby creating experience economy for the user. Similarly with the increase of dynamicity of the business the actual system use or adopters also increases (loop R8). The availability of component and proper application development helps the user in easy use of the health application which in turn influences the perceived usefulness of the user and their actual
IP T
system use (loop R9). Moreover, with the availability of the mHealth components helps the service
providers in maintaining price competitiveness which helps them to decide whether to release the
SC R
mHealth device or not (loop R10).
The balancing feedback loops B1, B2, and B3 corresponds to the privacy of the patient data in creating a degree of social influence or attitude towards using or intend to use as most of the users
N
5. Conclusion and Policy Implications
U
are concerned about these factors for system use.
A
This study addresses the sentiment or opinion of the people involved in the health service delivery,
M
as it is currently facing a lot of challenges in maintaining and delivering health service at an
TE D
affordable cost. The study considers peoples’ perceptions or feeling about the use of mHealth applications viz. fitness app, diabetes app, meditation app, and cancer app for health service delivery. By performing the sentiment analysis for the tweets, the result shows that mHealth
EP
applications for fitness, diabetes, and meditation app appears positive for tweets polarity and emotion values. Tweets also indicate joyful and satisfaction among the people in using these apps
CC
than compared to the cancer applications. Based on the positive, negative, and neutral values obtained through the sentiment analysis, a system thinking approach was also followed, by
A
drawing a CLD for identifying and understanding the influence of variables involved in making a mHealth application accessible and adaptable to the people. These findings help the healthcare practitioners and government in bringing upon important policy
implications and recommendations. It help the practitioners to get feedback about their health applications so that they can operate and modify them by regularly updating the technology databases with proper diet plan, calories meter and fitness videos and meditation tunes, or help the customer/patient in delivering health care
IP T
services from scheduling appointments to medication reminders including treatments and
processing of health records. This creates value for achieving mindfulness and satisfaction for the
SC R
patients and the customers. Furthermore, these data help the company to identify their current
market share with respect to their competitors which helps them in devising necessary policies and strategies accordingly. By focusing on the themes related to customer perceptions and expectations
U
along with its possible causal relationships of the system helps the company in increasing the
N
visibility of their health applications and returns to their stakeholders.
A
Findings help the government in identifying and communicating medical practitioners regarding
M
immunization or health-related programs so that they can plan for the medical training and visits
TE D
for improving treatment and care. It also assists the government in initiating and designing different insurance and health policy schemes for the benefit of the people particularly in the regions where access to health care is quite less.
EP
In future work, we intend to explore the perception and experience of the people in using mobile health applications such as LoseIt or mySugr Diabetes Logbook which help the application
CC
developers or service providers to provide better quality service. Moreover, causal loop diagram can be transformed to stock and flow diagram to measure the dynamics of different variables
A
influencing the accessibility and acceptability of the mobile health applications.
Author Contributions All authors have contributed to the writing of this research paper.
Acknowledgment This study did not receive any specific grant from funding agencies in the public, commercial, or
IP T
not-for-profit sectors. References
R. Ghosh, A. Datta, K. Lahiri, International OR Healthcare in emerging countries,
SC R
[1]
Challenges and opportunities for operations research, ORMS-Today (2013). [2]
Vital Wave Consulting, mHealth for Development: The Opportunity of Mobile Technology
first ed., Springer, New York, 2006.
A. Chib, M. H. van Velthoven, J. Car, mHealth adoption in low-resource environments: a
M
[4]
N
R. Istepanian, S. Laxminarayan, C. Pattichis, M-Health: emerging mobile health systems,
A
[3]
U
for Healthcare in the Developing World, 46 (2009) 1–70.
TE D
review of the use of mobile healthcare in developing countries, Journal of Health Communication, 20 (2015) 4-34. [5]
A. Gautam, Top Microblogging sites list with high pr and best 2016. My WP Tips.
[6]
EP
http://mywptips.com/top-microblogging-sites-list/ (accessed 3 May 2017). A. Agarwal, B. Xie, I. Vovsha, O. Rambow, R. Passonneau, Sentiment analysis of twitter
CC
data, In Proceedings of the Workshop on Languages in Social Media, Association for Computational Linguistics, USA, 2011, 30-38. P. Germanakos, C. Mourlas, G. Samaras, A mobile agent approach for ubiquitous and
A
[7]
personalized eHealth information systems. In Proceedings of the Workshop on ‘Personalization for e-Health ‘of the 10th International Conference on User Modeling (UM'05), Edinburgh, 2005, 67-70.
[8]
R. Evans, IMS Institute for Healthcare Informatics, Nursing Standard, 28 (2014) 30-30.
[9]
S. T. Liaw, Patient and general practitioner perceptions of patient-held health records, Family Practice, 10 (1993) 406-415.
[10] P.A. Abbott, J. Foster, H. de Fatima Marin, P.C. Dykes, P. C, Complexity and the science of
IP T
implementation in health IT—knowledge gaps and future visions, International Journal of Medical Informatics, 83 (2014) 12-22.
SC R
[11] Y. Jiang, S. M. Sereika, A. D. Dabbs, S. M. Handler, E. A. Schlenk, Using mobile health
technology to deliver decision support for self-monitoring after lung transplantation, International Journal of Medical Informatics, 94 (2016) 164-171.
U
[12] M. Siddiqui, M. Y. Islam, B. A. I. Mufti, N. Khan, M. S. Farooq, M. G. Muhammad, M.
N
Osama, D. Kherani, A. M. Kazi, Assessing acceptability of hypertensive/diabetic patients
A
towards mobile health based behavioral interventions in Pakistan: a pilot study, International
M
Journal of Medical Informatics, 84 (2015) 950-955.
TE D
[13] J. D. Sterman, Business Dynamics: systems thinking and modeling for a complex world, Irwin-McGraw-Hill, New York, 2000. [14] T. Binder, A. Vox, S. Belyazid, H. Haraldsson, M. Svensson, Developing system dynamics
EP
models from causal loop diagrams, In Proceedings of the 22nd International Conference of the System Dynamic Society, England, 2004, 1-21.
CC
[15] N. B. Lassen, R. Madsen, R. Vatrapu, Predicting iPhone sales from iPhone tweets. 2014 IEEE 18th International Enterprise Distributed Object Computing Conference (EDOC)
A
(2014) 81-90.
[16] P. Sedarati, A. Baktash, Smart glasses adoption in smart tourism destination: A conceptual model. In ENTER2017 Conference on Information and Communication Technologies in
Tourism. Rome, Italy. http://ertr.tamu.edu/content/issues/enter-2017-volume-8-researchnotes/ (accessed 13 June 2017) [17] S. Kanungo, Understanding the dynamics of information system use, In 21st International Conference of the System Dynamics Society, New York City, USA, 2003, 1-18.
IP T
[18] M. Mutingi, System dynamics of information technology adoption in a complex environment, International Journal of Industrial and Systems Engineering, 17 (2014) 78-96.
SC R
[19] D. Fisher, J. Norvell, S. Sonka, M. Nelson, Understanding technology adoption through
system dynamics modeling: implications for agribusiness management, The International Food And Agribusiness Management Review, 3 (2003) 281-296.
U
[20] Y. Chen, Understanding technology adoption through system dynamics approach: A case
N
study of RFID technology, 2011 IFIP 9th International Conference on Embedded and
A
Ubiquitous Computing (EUC), IEEE, Melbourne, Australia, 2011, 366-371.
M
[21] R. Wong, S. Y. Sheng, A system dynamics approach to the study of word-of-mouth and its
TE D
effect in an online environment, Technology Innovation Management Review, (2011) 4248.
[22] Y. Liu, H. Li, V. Kostakos, J. Goncalves, S. Hosio, F. Hu, An empirical investigation of
EP
mobile government adoption in rural China: A case study in Zhejiang province, Government Information Quarterly, 31 (2014) 432-442.
CC
[23] J. Cho, M. Quinlan, D. Park, G. Noh, Determinants of Adoption of Smartphone Health Apps among College Students, American Journal of Health Behavior, 38 (2014) 860-870.
A
[24] B. Liu, Sentiment analysis and subjectivity, In Handbook of Natural Language Processing, second ed., Chapman and Hall/CRC, 2010, 627-666.
[25] D. Cummings, H. Oh, N. Wang, Who needs polls? gauging public opinion from twitter data, 2010, Unpublished manuscript. [26] Johnson,
G.
Theories
of
Emotion
|
Internet
Encyclopedia
of
Philosophy.
http://www.iep.utm.edu/emotion/ (accessed 20 July 2017).
IP T
[27] A. Maslow, K. J. Lewis, Maslow's hierarchy of needs, Salenger Incorporated, 14 (1987).
[28] C. P. Alderfer, An empirical test of a new theory of human needs, Organizational Behavior
SC R
and Human Performance, 4 (1969) 142-175.
[29] E. A. Fagenson, Mentoring—Who needs it? A comparison of protégés' and nonprotégés' needs for power, achievement, affiliation, and autonomy, Journal of Vocational Behavior,
U
41 (1992) 48-60.
N
[30] J. Panksepp, Affective neuroscience: The foundations of human and animal emotions.
A
Oxford university press, 2004.
TE D
2nd ed., 2000, 91-115.
M
[31] L. Cosmides, J. Tooby, Evolutionary psychology and the emotions. Handbook of emotions,
[32] R. M. Nesse, Evolutionary explanations of emotions. Human nature, 1 (1990) 261-289. [33] R. Plutchik, Emotions: A general psychoevolutionary theory, Approaches to emotion,
EP
psychology press, 1984, 197-219.
[34] P. E. Griffiths, Is emotion a natural kind? In R. C. Solomon (Ed.), Thinking about feeling:
CC
Contemporary philosophers on emotions, Oxford university press, New York, 2004, 233– 249.
A
[35] C. Lutz, Unnatural emotions: Everyday sentiments on a Micronesian atoll & their challenge to western theory, Chicago: university of Chicago press, 1988.
[36] Harré, R, The social construction of emotions, Basil Blackwell, 1986.
[37] D. Centeno, Nurtured Emotional Evolution among the Ifaluk. Apathy and Prophecies. https://astrowipe.wordpress.com/2009/07/03/nutured-emotional-evolution-among-theifaluk/ 2009, (accessed 20 July 2017). [38] S. Rauch, E. Foa, Emotional Processing Theory (EPT) and Exposure Therapy for PTSD.
IP T
Journal of Contemporary Psychotherapy. 36 (2006) 61-65.
[39] M. Nussbaum, Emotions as judgements of value and importance. In R. C. Solomon (Ed.),
SC R
Thinking about feeling: Contemporary philosophers on emotions, Oxford university press, 2004, 183–199.
[40] I. J. Roseman, Cognitive determinants of emotion: A structural theory, Review of Personality
U
& Social Psychology, 1984.
N
[41] D. G. Myers, Myers' psychology for AP, Macmillan, 2010.
A
[42] V. S. Pagolu, K. N. R. Challa, G. Panda, B. Majhi, Sentiment Analysis of Twitter Data for
M
Predicting Stock Market Movements, 2016 International Conference on Signal Processing,
1345-1350.
TE D
Communication, Power and Embedded System (SCOPES), IEEE, Odisha, India, 2016,
[43] M. Rastegar-Mojarad, Z. Ye, D. Wall, N. Murali, S. Lin, Collecting and Analyzing Patient
EP
Experiences of Health Care From Social Media, JMIR Research Protocols, 4 (2015) e78. [44] C. Wong, M. Sap, A. Schwartz, R. Town, T. Baker, L. Ungar, R. Merchant, Twitter
CC
Sentiment Predicts Affordable Care Act Marketplace Enrollment, Journal of Medical Internet Research. 17 (2015) e5, 1 1-8.
A
[45] M. J. Paul, M. Dredze, You are what you Tweet: Analyzing Twitter for public health, Fifth International AAAI Conference on Weblogs and Social Media, AAAI Publications, 20 (2011) 265-272.
[46] I. Korkontzelos, A. Nikfarjam, M. Shardlow, A. Sarker, S. Ananiadou, G. H. Gonzalez, Analysis of the effect of sentiment analysis on extracting adverse drug reactions from tweets and forum posts, Journal of Biomedical Informatics. 62 (2016) 148-158. [47] M. Salathé, S. Khandelwal, Assessing vaccination sentiments with online social media:
IP T
implications for infectious disease dynamics and control, PLoS Computational Biology, 7 (2011), e1002199.
SC R
[48] F. D. Davis, R. P. Bagozzi, P. R. Warshaw, User acceptance of computer technology: a
TE D
M
A
N
U
comparison of two theoretical models. Management science, 35 (1989) 982-1003.
A
CC
EP
Figures
IP T SC R U N
A
CC
EP
TE D
M
A
Figure 1: Flow diagram for extracting tweets (single column fitting image)
Figure 2a: Bigram for fitness applications (single column fitting image)
IP T SC R U
800
741
A
Sentiment analysis of tweets about fitness app (classification by polarity)
M
600 500 400 300
TE D
Number of tweets
700
N
Figure 2b: Word cloud for fitness applications (2-column fitting image)
200 100 0
EP
Positive
202 57 Negative
Neutral
A
CC
Figure 3a: Sentiment analysis for fitness app based on polarity (single column fitting image)
800
742
Sentiment analysis of tweets about fitness app (classification by emotion)
700
500 400 300 200
132 71
100
16
16
Fear
Anger
Unknown
Joy
Sadness
16
SC R
0
IP T
Number of tweets
600
Surprise
EP
TE D
M
A
N
U
Figure 3b: Sentiment analysis for fitness app based on emotion (2-column fitting image)
A
CC
Figure 4a: Bigram for diabetes applications (single column fitting image)
7
Disgust
IP T SC R U
259
M
200 150
TE D
Number of tweets
250
100
EP
50 0
Sentiment analysis of tweets about diabetes app (classification by polarity)
A
300
N
Figure 4b: Word cloud for diabetes applications (2-column fitting image)
Positive
56
50
Negative
Neutral
A
CC
Figure 5a: Sentiment analysis for diabetes app based on polarity (single column fitting image)
350
Sentiment analysis of tweets about diabetes app (classification by emotion)
300
250 200
100 50
28
23
8
0 Unknown
Sadness
Joy
Anger
4
IP T
150
SC R
Number of tweets
300
Fear
2
Surprise
EP
TE D
M
A
N
U
Figure 5b: Sentiment analysis for diabetes app based on emotion (2-column fitting image)
A
CC
Figure 6a: Bigram for meditation applications (single column fitting image)
IP T SC R
900
810
Sentiment analysis of tweets about meditation app (classification by polarity)
N
800
U
Figure 6b: Word cloud for meditation applications (2-column fitting image)
A
600
M
500 400 300
TE D
Number of tweets
700
200 100 0
EP
Positive
112
Negative
78
Neutral
A
CC
Figure 7a: Sentiment analysis for meditation app based on polarity (single column fitting image)
1000 900
Sentiment analysis of tweets about meditation app (classification by emotion)
879
700 600 500 400 300 200 81 18
8
Disgust
Sadness
0 Unknown
Joy
8
SC R
100
IP T
Number of tweets
800
Surprise
6
Anger
EP
TE D
M
A
N
U
Figure 7b: Sentiment analysis for meditation app based on emotion (2- column fitting image)
A
CC
Figure 8a: Bigram for cancer application (single column fitting image)
IP T SC R
Sentiment analysis of tweets about cancer app (classification by polarity)
N
600
U
Figure 8b: Word cloud for cancer application (2-column fitting image)
491
A M
400
300
TE D
Number of tweets
500
200
EP
100
0
172
164
Positive
Negative
Neutral
A
CC
Figure 9a: Sentiment analysis for cancer app based on polarity (single column fitting image)
700
654
Sentiment analysis of tweets about cancer app (classification by emotion)
500 400
200 100
60
43
32
Sadness
Fear
0 Unknown
Joy
20
IP T
300
SC R
Number of tweets
600
Anger
TE D
M
A
N
U
Figure 9b: Sentiment analysis for cancer app based on emotion (2-column fitting image)
A
CC
EP
Figure 10a: Boxplot for mobile health applications (2-column fitting image)
18
Surprise
IP T SC R U
CC
EP
TE D
M
A
N
Figure 10b: Sentiment scores for mHealth applications (2-column fitting image)
Figure 11: Causal loop diagram showing the relationship between the potential adopters, semi-
A
adopters, and adopters of mHealth system
IP T SC R U N A
M
Figure 12: Causal loop diagram for perceived usefulness of mHealth system
TE D
Table 1: Total number of tweets extracted mHealth apps Tweets requested Tweets obtained 1000
1000
Diabetes app
1000
365
Meditation app
1000
1000
Cancer app
1000
827
CC
EP
Fitness app
Table 2: Tweets for fitness applications Tweets
Emotion
Polarity
A
Jenn at jensfunfinds says her favorite thing about japrasportpace is the Unknown Positive fitness test that comes with the app Officially logged my meals into my fitness pal app for days in a row Fear highly recommend it to everyone takes minutes
Positive
NL sports coming soon to a sports team near you sport coaching training Joy
Positive
apps tech fitness Can I put on my fitness app that I did in of squats because I squatted Anger
Negative
down to get my hidden treats and I was too high to get back up Sadness
Neutral
Discover why today is an amazing day to get started on that fitness or Surprise
Neutral
IP T
Get a free day pass to best fitness activate now with my special code
business goal
Genuinely paying pound a month for a fitness app if this won’t motivate Disgust
SC R
me nothing will be tired of feeling crap Table 3: Tweets for diabetes applications Tweets
Emotion
Neutral
Polarity
U
A mobile app for the self-management of type diabetes among Unknown Positive
N
adolescent a randomized controlled trial
Wear your medical alert bracelet or update the health on your Fear
A
smartphone, so others know you have type diabetes
or deal with diabetes very cool
M
Glad to hear I saw this Israeli app that can personalize diets to prevent Joy
Anger
TE D
He knows how to scan the QR code and get the diabetes risk score app
At a time when insulin amp other diabetes drugs are soaring Amazon Surprise
Positive
Positive
Negative Negative
creates a stupid distraction diabetes app contest duped again, folks Discover how breakthrough diet sensor nutrition app helps in managing Sadness
Neutral
EP
diabetes weight loss and improves fitness
CC
Table 4: Tweets for meditation applications Tweets
Emotion
Polarity
Meditation app headspace hires a new chief business officer tech crunch Unknown Positive Positive
A
Is the best app in fall asleep to the meditation sleep times in likes 25 Joy seconds amazing How to deal with pesky thoughts in meditation success
Anger
Headspace’s revamped app helps busy people turn meditation into habit Disgust wellness
Negative Negative
Heal your body mind and soul with these amazing meditation on iTunes Surprise
Neutral
for iPhone Wow just discovered meditation apps is there such a thing as a get off Sadness
Neutral
your butt amp write app cos hearing katching amwriting Table 5: Tweets about cancer applications Emotion
Polarity
IP T
Tweets
Sharktankaucancer touches us all so I really want this app to succeed Unknown Positive
SC R
amp make life impact changes Our nutrition app can help you find the optimal diet for any stage of your Fear cancer journey
Great to see innovative thinkers in the cancer care space sharktankau
Joy
U
Cancer app rides high on emotion surfer creates comic relief for patient Anger
N
brain tumor
Get the free mindfulness app for those living with cancer and their Sadness
A
caregivers
more than k people in countries
Positive Negative
Negative
Neutral
M
Cancer aid is the top cancer app in Australia UK amp the US used by Surprise
Positive
mHealth apps
TE D
Table 6: Boxplot values for mobile health apps median
Quartile values
Adjacent values
upper
lower
upper
lower
-1
-1
-
+1
-1
Diabetes
+1
-
+1
+2
-1
Meditation
0
+1
-
+2
-1
Cancer
0
+1
-
+2
-1
A
CC
EP
Fitness
S1386-5056(18)30119-9 https://doi.org/10.1016/j.ijmedinf.2018.02.016 IJB 3666
To appear in:
International Journal of Medical Informatics
Received date: Revised date: Accepted date:
13-9-2017 20-2-2018 22-2-2018
Please cite this article as: Rajesh R.Pai, Sreejith Alathur, Assessing mobile health applications with twitter analytics, International Journal of Medical Informatics https://doi.org/10.1016/j.ijmedinf.2018.02.016 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ASSESSING MOBILE HEALTH APPLICATIONS WITH TWITTER ANALYTICS
Author names and affiliations: Rajesh R. Paia,*, Sreejith Alathurb a,b
School of Management, National Institute of Technology Karnataka, Surathkal,
IP T
India
SC R
[email protected]
*Corresponding author:
U
Rajesh R. Pai
N
School of Management, National Institute of Technology Karnataka, Surathkal,
A
India
M
[email protected]
TE D
Highlights
Sentiment analysis was performed on the Twitter data using the keywords viz. fitness apps, diabetes app, meditation app, and cancer app representing the types of mobile health applications.
Perception and usage experience of mobile health applications among the citizens has been
EP
studied and identified that except for cancer application there exists a positive polarity
CC
towards the fitness, diabetes and meditation application among the users.
System thinking approach followed with the help of causal loop diagram for identifying
A
the relationships between the variables influencing the accessibility and acceptability of mobile health application among the citizens, healthcare, patients and application developers which help them in understanding the dynamics involved in system.
Introduction: Advancement in the field of information technology and rise in the use of Internet has changed the lives of people by enabling various services online. In recent times, healthcare sector which faces its service delivery challenges started promoting and using mobile health applications with the intention of cutting down the cost making it accessible and affordable to the
IP T
people.
Objectives: The objective of the study is to perform sentiment analysis using the Twitter data
SC R
which measures the perception and use of various mobile health applications among the citizens.
Methods: The methodology followed in this research is qualitative with the data extracted from a social networking site “Twitter” through a tool RStudio. This tool with the help of Twitter
U
Application Programming Interface requested one thousand tweets each for four different phrases
N
of mobile health applications (apps) such as “fitness app”, “diabetes app”, “meditation app”, and
M
emotions were measured.
A
“cancer app”. Depending on the tweets, sentiment analysis was carried out, and its polarity and
TE D
Results: Except for cancer app there exists a positive polarity towards the fitness, diabetes, and meditation apps among the users. Following a system thinking approach for our results, this paper also explains the causal relationships between the accessibility and acceptability of mobile health
EP
applications which helps the healthcare facility and the application developers in understanding and analyzing the dynamics involved the adopting a new system or modifying an existing one.
CC
Keywords: Mobile Health, Sentiment Analysis, Twitter Analytics, Causal Loop Diagram, Technology Adoption Model
A
1. Introduction Health is considered to be the most important component in the lives of every human being. Given such a priority, maintaining good health with healthy diet, moderate exercises, and regular medical
checkups are the greatest challenges which people are facing currently in this busy world. Moreover, with the large geographical area and population level, the dichotomy in demand for the healthcare and lack of health insurance coverage resulted in increased cost of the healthcare services. Consequently, this has created an impact on providing better healthcare and treatment
IP T
services across rural and urban cities and eventually has become a major concern for many
governments worldwide [1]. To reduce the disparity of population level (5,300 million) and
SC R
healthcare (11 million), various governments and healthcare sectors, are promoting awareness about mobile technology use in healthcare services for the people living in the rural and urban
regions [2]. The use of mobile technology in communicating healthcare practices to the citizens
U
through health applications (apps) are often termed as mobile health (mHealth). It was first defined
N
by Robert Istepanian as “emerging mobile communications and network technologies for
A
healthcare systems” [3]. This technology is considered to be the backbone of electronic health
M
(eHealth) due to its ability to provide healthcare access to the people living in rural areas. These
TE D
apps act as interphase between the people and the physician/doctors for consultations and monitoring through videos or messages facilitating them in maintaining health. Mobile health has been emerged due to rapid growth in the use of mobile technologies and its
EP
connectivity to access health-related information. Thus transforming the present health service delivery practices to more of non-physical interaction between Healthcare Providers (HCP) and
CC
patient/user [4]. This is required especially for the people living in a low-resource setting wherein access to healthcare facilities and services are often minimal. To tackle this, government and
A
technology entrepreneurs are promoting various mHealth apps with the intention of reducing healthcare service cost making it convenient for emergencies and general health services. Even though some people are using mHealth apps on a regular basis for maintaining their health,
the data results or evidence regarding health outcomes are still not clear. In such cases, microblogging sites due to the nature and accessibility of data availability can be used to collect peoples’ opinions about these apps. These data along with consumer health informatics can be used for generating real-time information from user posts relating to the products, service,
IP T
individual, events, and place which helps the user in building trust towards the use of mHealth
apps. Currently, there are various microblogging sites such as Twitter, Google+, Facebook, etc.
SC R
however, Twitter and the Facebook are the most commonly used.
Therefore, the current study uses Twitter data for analyzing the people’s sentiment, as it is considered to be the most popular page rank microblogging sites of the year 2016 depending on
U
the number of users and page views [5]. A user here on this site can post a message or tweet relating
N
to the product or an event with a word limit of 140 characters. Moreover, it can also be used as an
A
online news service portal through which individuals can view discussions which are happening
M
across the globe.
TE D
Many times, these tweets play an important role in manufacturing and service organizations to study the user's reaction or responses about its products or services which helps them in making decisions. However, the greatest challenge is to organize these data so as to transform it into useful
EP
facts and draw conclusions [6]. Thus the analysis performed in this study extracts tweets and analyzes user’s sentiments or emotions by classifying them into positive, negative or neutral using
CC
a tool called as RStudio through different R-codes. The current study makes an attempt to access the sentiment score for different mHealth apps
A
through a social networking site called Twitter, and thereupon causal relationships have been established to identify the interrelationships between the factors.
2. Literature Review Mobile technologies are growing dramatically in recent years due to the advancement in the field of information and communication technologies. This has changed the lifestyle of the people through different mobile applications (apps) which in turn changed the way of communication and
IP T
living. Since then, mobile technologies are in continuous use for delivering a large range of
services starting from calling, messaging, etc. up to its utility in banking sectors due to the features
SC R
such as speed of accessibility and /or mobility. Currently, it has entered into healthcare sectors
with different features coupled with computer-supported technologies and medicine. In healthcare,
doctors for monitoring patient and delivering care [7].
U
these apps find its use in collecting community health data and its delivery to practitioners, and
N
According to the study conducted by various research associations, there are more than 165,000
A
mobile health (mHealth) apps available across different mobile app stores [8]. Most of these apps
M
finds its use in maintaining Health and Fitness (such as LoseIt, Google fitness, Fitbit, etc.), Diet
TE D
Companion Apps (Health and Nutrition, Calorie counter etc.), Stress Relief Apps (Simple Yoga, Relax meditation: Sleep sounds, etc.), Diabetes apps (mySugr Diabetes Logbook, Glucose Buddy: Diabetes Log, etc.) etc. Concerning patient health record, it was identified that doctors and patients
EP
have different attitudes and expectations which found to have significant sociodemographic, educational and attitude correlation with patient-held health records [9]. Additionally, electronic
CC
and other Information Technology (IT) system adoption in a complex healthcare environment slows- down process as it doesn’t have one size fits all devices and requires consideration of many
A
healthcare and technology dimension [10]. Jiang et al. [11], studied the response of lung transplant recipients towards a mHealth app ‘Pocket PATH®’ which was designed to provide automatic feedback messages detecting and reporting
critical values of health and found to be appropriate among the recipients. Likewise, when the app is used for hypertensive and diabetic patients, a phone call or SMS reminders was an acceptable and favorable option for managing the chronic diseases [12]. However, research is still in process with various techno-entrepreneurs and academicians granting maximum access and support to the
IP T
people in enhancing their health behaviors.
Many researchers have explained the potential of using mHealth in supporting health workers and
SC R
citizens of the low and medium income settings. However, the evidence base with respect to
documentation, designs, and reported outcomes are weak and inconsistent. Hence, its utility can be measured by studying the complexity of the system involved in the mHealth adoption by
U
understanding the nature of sentiments people carry during its purchase and use. Thus, this section
A
(called sentiment analysis) for the Twitter data.
N
discusses the use of system dynamics in the mHealth adoption followed by social media analytics
M
2.1. System Dynamics and Mobile Health Adoption
TE D
Sentiment score executed from the Twitter sentiment analysis can be used with the set of differential equation to study the interactions between the dependent and independent variables of a system which is defined. In addition, system thinking approach can also be followed in the form
EP
of causal loop diagram (CLD) so as to measure the factors influencing mHealth adoption and accessibility among the people. These CLD’s are represented with different feedback structures of
CC
the systems by capturing individuals or teams mental model or communicating the feedbacks which are highly responsive causing dynamics [13]. It can also be used in standard system dynamic
A
practices for connecting the simulation analysis and modeling to study the behavior of the system over time [14]. System Dynamics mathematics with tweets sentiment values can be used to predict smartphone
sales helping the company to analyze customer perception about their offering [15]. Sedarti and Batash [16], used CLD to understand human behavior involved in the dynamics of wearable smart devices use and identified social influence as a significant factor for the device acceptance. The dynamics of the user behavior was also studied by extending the hypothesis and incorporating
IP T
various feedbacks leading from Information System (IS) use to perceived ease of use on one side
and productivity to IS-related work on the other. These feedbacks help managers in devising
SC R
strategies during system changes or introduction of new systems [17].
Consequently, the managers require in-depth knowledge about technology adoption and their behavior when influenced by adopters and inhibitors depending upon design and technology
U
evaluation policies. Therefore concluded that peak adoption time delay and peak adoption rate
N
could be affected by the adoption inhibitors and the diffusion process of technology [18]. This is
A
because of potential and current technology adopters who are unaware of the benefits of
M
technology adoption which they learn through information feedback present in the system. In such
TE D
cases, system dynamics modeling can be used to overcome facilitating factors of time and adoption [19].
Chen [20], identified that organization would accept technology based on management attitude
EP
and perception which can be made favorable through effective promotions and word-of-mouth. Likewise, customers also consider the return on investment, time and effort from that technology.
CC
Wong and Sheng [21] identified that word-of-mouth with moderating factors such as word-ofmouth valence, trust, and risk attitude reduces uncertainty involved in purchasing or consuming
A
any products and services. Liu et al. [22], investigated the adoption of mobile government among the rural population and identified that young males having knowledge about policies have positive perception and become
potential adopters of the technology. They can later be transformed into adopters through wordof-mouth, perceived ease of use, usefulness, image and social influence. Moreover, technology adoption is also dependent on the factors such as subjective norm, health consciousness, health information orientation, electronic health literacy, internet health information use efficacy,
IP T
perceived usefulness, perceived ease of use, and behavioral intention to use when young student populations are considered [23].
SC R
2.2. Social Media Analytics and Mobile Health
Social media analytics came into the visibility due to availability and accessibility of a large amount of data which organizations across the world are finding it difficult to manage and derive
U
a meaningful solution. Therefore, various analytics has been identified among them sentiment
N
analysis is the one which assists organizations in establishing data relationships.
A
Sentiment analysis or opinion mining focuses on the computational study of opinions, sentiment,
M
and emotions which are expressed in the form of text [24]. In other words it refers to the processing
TE D
of natural language, text analysis, and biometrics to identify, extract, quantify, study the affective states, and subjective information about a product/service, or an event from the perspective of the customer through reviews, survey responses obtained from the social media or blogs [25]. These
EP
responses are helpful for the organizations to measure the sentiment of an individual having towards their products or services which assists them in production and marketing.
CC
In psychology, there exist different theories which explain peoples’ need and emotions towards the mHealth app use for health activities. But then most of these depend on how we perceive our
A
social environment [26] and this may not be constant until people get experiences over time. Hence, the following section reviews select literature which drives peoples’ needs and emotions for the use of mHealth apps.
Theories of needs and motivation: According to Abraham H. Maslow, people will be motivated to use products only when their needs are unmet or unsatisfied [27]. These unsatisfied need depends upon satisfaction which influences the behavior driving them to progress upwards in the hierarchy level as classified by Alderfer’s in
IP T
ERG theory of Existence (need physiological health and safety), Relatedness (need for
interpersonal connections, social status, and recognition) and Growth (need for personal
SC R
development) [28]. These motivations can also arise because of needs associated with achievement, power, and affiliation as David McClelland explained in his three needs theory [29].
Therefore, in the case of mHealth app use the primary reason being the disparity in the level of
U
population and healthcare accessibility resulted in problems related to the delivery of good quality
N
healthcare services. Since these needs are unmet, research associations, healthcare units, and the
TE D
Theories of emotions:
M
information’s over mobile phones.
A
government started promoting health apps such that a number of people can have access to health
Emotions are the feeling which a person expresses depending on circumstances or environment. It can be any conscious experiences characterized by intense mental actions resulting in high level
EP
of joy or disappointment [30]. In evolutionary theories, emotions such as fear, jealousy, anxiety, anger, desire, love, etc. had evolved from the early hominids, using certain adaptations such as
CC
incorporation, rejection, destruction, protection, reproduction, reintegration, orientation, and exploration, which are being shared by animals, resulting in the processing of certain traits due to
A
their shared ancestry [26, 31-34]. Social and cultural theories of emotions consider emotions which starts with an idea and a product of societies and culture acquired by the individuals through experience [26]. This theory also
addresses emotion words used in different languages for representing a particular behavior or situation [35, 36]. Example: in Japanese, the word ‘Fago’ describes maturity level of the person like childishness, selfishness, etc. and ‘Song’ refers to unacceptable behavior [37]. Further, emotion process starts from the perception of stimulus then triggers a response of the body and
IP T
finally body responses in terms of change in blood pressure, heartbeat, facial expression, etc. i.e.,
conceptualization of anxiety disorders [26, 38]. For example, anger because of insult and joy
SC R
because of good grades in exams.
In judgment theories of emotion process, emotions such as surprise, hope, fear, joy, relief, discomfort, disgust, etc. are considered to have a basic judgment about ourselves and our place in
U
the world based on a projection of ethics and values, structures and mythologies. However, the
N
cognitive theories signify the way the individuals evaluate the situation determines the emotions
A
[39, 40].
James-Lange Theory: It proposes that the people will experience the emotions when they
TE D
M
Some of these emotion theories as defined and distinguished by David G. Myers [41] were:
perceive things. Example: When we feel sad, we cry, likewise when we feel embarrassed we blush.
Cannon-Bard Theory: This theory suggests that people will experience emotions and
EP
physiological arousal at the same time, which means that emotions and arousal are
Schachter-Singer “Two-factor” Theory: It says people will experience emotions which depend
A
CC
simultaneous. Example: We blush and feel embarrassed at the same time simultaneously.
upon physiological arousal and its cognitive interpretation. Example: The label which people give for the same physiological arousal at different instances and context like fear can become arousal depending on the environment.
Robert Zajonc, Joseph LeDoux, and Richard Lazarus: Emotions without Awareness/Cognition Theory: This theory states that some reactions which are emotional develop on a low road through brain skipping the thought of consciousness. Example: In the forest, we respond to the sound before assessing it.
IP T
Based on these theories many researchers across the world have studied the sentiments or emotions of the people they followed while writing books, blogs, social networking sites, etc. Among them,
SC R
some of their findings have been explained in the literature below.
Pagolu et al. [42], found that there is a strong correlation between the stock price and public
U
opinions by analyzing Twitter data. The Twitter data is also analyzed based on volume, generic sentiment, and language model for predicting the movement of the public opinion measured
N
through presidential approval rating, economic confidence, and the generic Congressional ballot
A
[25]. Lassen et al. [15], used sentiment analysis by developing a linear regression model for
M
predicting the sales of the iPhone from its tweets which are found to have a strong correlation.
TE D
Twitter can also be used to analyze the sentiment of the people about particular healthcare or public health measures. Rastegar-Mojarad et al. [43], used sentiment analysis techniques and Hadoop to analyze the patient experience of healthcare and concluded that longer the patient reviews poorer
EP
the reviews of the facility. Twitter analysis can also be used to measure the real-time barometer of public on Affordable Care Act (ACA) and identified that there exists a positive correlation between
CC
the sentiment and marketplace enrollment [44]. Paul and Dredze [45], correlated influenza rates
A
with the Twitter data and indicated that for a particular symptom of infections or physical injuries the intended usage of medicine can be mentioned. It can also be used for extracting and investigating the effect of adverse drug reactions helping them in improving drug and creating popularity in the social media and other health forums [46]. Salathe´ and Khandelwal [47],
investigated the spatiotemporal sentiments for a novel vaccine which can be used to measure the intuition about that vaccine. Hence they concluded that groups of negative vaccine sentiments leads to unprotected individual and increases disease outbreaks. Considering these literature, this research paper uses Twitter data to study the people’s perceptions and experiences towards the
IP T
mHealth app use. 3. Methodology
SC R
The methodology followed in this research is qualitative with the data extracted from a social
networking site “Twitter” using RStudio. The framework for the tweets extraction is represented
U
in figure 1. 3.1. Dataset
N
RStudio tool is used to extract Twitter access using ROAuth and web API (Application
A
Programming Interface) and requested 1000 tweets each for four different phrases of mHealth
M
apps, namely fitness app, diabetes app, meditation app, and cancer app. Tweets were extracted for
TE D
latest 8 days starting from 08 June 2017 to 15 June 2017, and sentiment analysis was performed to measure its polarity and emotions. 3.2. Sentiment analysis of tweets
EP
This paper uses word cloud, and bigram for variable frequency identification found on the Twitter website. From this platform, using R codes tweets with keywords viz. fitness app, diabetes app,
CC
meditation app, and cancer app were extracted, and sentiment analysis was performed. Table 1,
A
indicates tweets requested and obtained for each of the mHealth apps: 3.3. Causal loop diagram for mHealth adoption Causal Loop Diagrams (CLD) visualizes the cause and effect relationships between variables involved in mHealth devices adoption among citizens. CLD’s are represented by polarities
(positive (+) and negative (-)) and delay. A positive polarity between ‘A’ and ‘B’ indicates as ‘A’ increases ‘B’ also increases and a negative polarity represents as variable ‘A’ increases, variable ‘B’ decreases in the opposite direction [13]. It is used in this study considering the variables of Technology Adoption Model (TAM) of Davis et al. [48]. This model considers perceived
for measuring the intention to use and actual system use or adopters.
SC R
4. Result and Analysis
IP T
usefulness, perceived ease of use, and attitude of using the technology as the major components
The RStudio requested 1000 tweets for which bigram, word cloud, and sentiment scores were
U
calculated under four different cases depending upon the type of mHealth apps. Case i: Twitter analytics for fitness applications
N
Figure 2a and 2b indicate that people using iPhones are aware of fitness apps which are new and
A
freely available and are discussing its importance in managing services at a clinic or for
M
maintaining a balanced diet (food diary, nutrition, etc.). Moreover, users may also consider if
TE D
suggested by clinics or hospitals such as Mayo clinic, etc. Figure 3a specifies that about 74% of the tweets have positive sentiments, indicating that people are seriously considering fitness apps as a tool for improving their health. It is supported by figure
EP
3b, indicating application users’ emotion as joyful (132 tweets) and higher when compared other emotion components.
CC
The following Table 2 below shows select tweets with its polarity and emotion values.
A
Case ii: Twitter analytics for diabetes applications Figure 4a and 4b indicate that people may be aware of diabetes mobile apps used for monitoring glucose level in the blood through one-touch digital health meters revealing the type of hypoglycemia (type 1 and type 2) among the people.
The total number of tweets for diabetes app extracted during the analysis period was only 365 numbers. This value appears to be less than fitness application and may improve over time as the positive sentiment tweets are 259 representing 26% of the total tweets extracted (figure 5a). The reason behind this may be because of awareness or interoperability issues between users and
IP T
doctors in establishing interphase resulted in sadness. Alternatively, people using this app frequently on a daily basis are found to be joyful in using the application (figure 5b).
SC R
The following Table 3 provides the analysis of tweets extracted based on emotion and polarity. Case iii: Twitter analytics for meditation applications
U
Figure 6a and 6b illustrate that people often talk about these applications due to its mindfulness
N
and deep sounds features which are binaural and available freely in mobile play stores. This
A
application helps the user to feel relaxed through their audios and improve concentration helping
M
them in brain development.
From the figure 7a, we can identify that there is a positive opinion about the meditation app among
TE D
people as the total number of positive sentiment is 810, which corresponds to 81% of the total tweets extracted. Results of the sentiment analysis also indicate that 8.1% of the people/user are joyful in their perception and use of this app (figure 7b). This signifies that meditation app is
EP
working fine and people are satisfied with it. Some of the select comments about this applications
CC
are shown in the table below. Case iv: Twitter analytics for cancer applications
A
Figure 8a and 8b show that people are discussing about cancer application available in different play stores but are much concerned about the survival benefits as it is paradoxical in nature with an example being ‘a cough turned out to be lung cancer via the app.' Many users suggest that use of mobile cancer application in the early stages of diagnosis by displaying some documented
symptoms can be helpful in improving and saving lives. For the cancer application, the number of positive sentiments is less compared to that of negative (figure 9a). This indicates that the perception and use of this application appear to be less as most of the people may prefer visiting the doctors personally than through smartphones for their regular
IP T
checkups. Though, there are more negative sentiment values people have expressed joyful (60
tweets) when compared with other emotional parameters (figure 9b). This result can be because of
SC R
features such as diet tracker, first aid process, finding the doctor and for booking appointments,
etc. which are available in this application. Table 5 below shows the examples of different tweets
U
extracted from various cancer applications.
N
Boxplot and sentiment scores for mobile health applications
A
The following figure (figure 10a and 10b) shows the boxplot indicating interquartile range values
M
and sentiment scores for four different mHealth applications.
TE D
Considering cancer app (figure 10a), we can identify the existence of positive skewness (median less than mean) in the negative region as the data are distributed below the median values of the boxplot. This indicates that people perception or use of cancer app is less than the average
EP
perception or use which is also identified in the figure 10b. Table 6 shows the boxplot for mHealth
A
CC
application indicating median, quartile and adjacent values.
Moreover, for diabetes app, the median is more than the mean (negative skewness) but in the positive region. These extreme positive sentiments are considered to be good for the company
because of the people’s positive perception towards the app use (figure 10a). This can also be identified from the figure 10b as more than 50% of the data falls onto the right side or positive side of the graph. Positive skewness is also seen even in the case of fitness and meditation app (figure 10a) with the
IP T
data in the positive region having almost the same sentiments value with progressive signs about
the apps as 50% of data falls on the positive side (figure 10b). This means that people realized the
SC R
importance and tended to use these applications to maintain their health which can, in turn, help the company to update their health applications to remain competitive.
Systems thinking approach for analyzing mobile health application adoption
N
feedback loops for understanding systems’ behavior.
U
From the figure 11 and figure 12, we can consider the presence of reinforcing and balancing
A
[Insert figure 11] (2-column fitting image)
M
Figure 11: Causal loop diagram showing the relationship between the potential adopters, semi-
TE D
adopters, and adopters of mHealth system
[Insert figure 12] (2-column fitting image)
EP
Figure 12: Causal loop diagram for perceived usefulness of mHealth system The first reinforcing loop R1, named as “benefits from the use of mHealth loop” connects the user
CC
and mHealth application willingness in increasing the completion rate. This means that more the number of mHealth users, more they satisfy with the application resulting in effective utilization
A
of mHealth system thus achieving desired health target. At the point of time when the user gets benefitted, their desire for mHealth also increases. This encourages healthcare providers or application developers to put efforts on the quality improvement of the information system which improves the overall system quality. When this increases, the user may feel that the system is
becoming more useful and easier to use which leads to dependent and willing to use. The second reinforcing loop R2, named as “users perception of quality and usage” basically explains, more the number of users more they will satisfy with the system and feels that the system is becoming more useful increasing the attitude of using health application more than they used
IP T
earlier.
The third reinforcement loop R3, named as the “learning loop” is developed based on user learning
SC R
which takes place when the user actually uses the system. In other words, more the number of users more the learning taking place and with electronic health (eHealth) literacy develops self-
efficacy or confidence in achieving targets easily thereby increasing attitude and willingness to
U
use the system.
N
“Market saturation loop (R4)” and “word of mouth loop (R5)” focuses on transforming new and
A
potential adopters to semi-adopters which later become the adopters of the system. The story
M
behind these loops represents, as the potential adopter of the system through proper promotional
TE D
take-ups (by creating awareness) improves the take-up rate which also increases along with the adopters contact through word of mouth take-up resulting from mHealth system/application adoption. This increase in the take-up rate reduces the number of potential adopters thereby
EP
reducing the adoption or use of the system. Moreover, the take-up rate is also influenced by the accessibility factors, i.e., IT infrastructure, technology literacy and socioeconomic dimensions
CC
(such as age, income, and education) relating to mHealth system or application. The reinforcement loops R6 and R7 represents the transformation of the users from semi-potential
A
adopters to adopters (or actual system use) through practice rate and learning on continuous use of the system. In a mHealth system, more the number of adopters or users more the ease in patient data collection, facilitating data process assisting healthcare providers in business dynamicity
thereby creating experience economy for the user. Similarly with the increase of dynamicity of the business the actual system use or adopters also increases (loop R8). The availability of component and proper application development helps the user in easy use of the health application which in turn influences the perceived usefulness of the user and their actual
IP T
system use (loop R9). Moreover, with the availability of the mHealth components helps the service
providers in maintaining price competitiveness which helps them to decide whether to release the
SC R
mHealth device or not (loop R10).
The balancing feedback loops B1, B2, and B3 corresponds to the privacy of the patient data in creating a degree of social influence or attitude towards using or intend to use as most of the users
N
5. Conclusion and Policy Implications
U
are concerned about these factors for system use.
A
This study addresses the sentiment or opinion of the people involved in the health service delivery,
M
as it is currently facing a lot of challenges in maintaining and delivering health service at an
TE D
affordable cost. The study considers peoples’ perceptions or feeling about the use of mHealth applications viz. fitness app, diabetes app, meditation app, and cancer app for health service delivery. By performing the sentiment analysis for the tweets, the result shows that mHealth
EP
applications for fitness, diabetes, and meditation app appears positive for tweets polarity and emotion values. Tweets also indicate joyful and satisfaction among the people in using these apps
CC
than compared to the cancer applications. Based on the positive, negative, and neutral values obtained through the sentiment analysis, a system thinking approach was also followed, by
A
drawing a CLD for identifying and understanding the influence of variables involved in making a mHealth application accessible and adaptable to the people. These findings help the healthcare practitioners and government in bringing upon important policy
implications and recommendations. It help the practitioners to get feedback about their health applications so that they can operate and modify them by regularly updating the technology databases with proper diet plan, calories meter and fitness videos and meditation tunes, or help the customer/patient in delivering health care
IP T
services from scheduling appointments to medication reminders including treatments and
processing of health records. This creates value for achieving mindfulness and satisfaction for the
SC R
patients and the customers. Furthermore, these data help the company to identify their current
market share with respect to their competitors which helps them in devising necessary policies and strategies accordingly. By focusing on the themes related to customer perceptions and expectations
U
along with its possible causal relationships of the system helps the company in increasing the
N
visibility of their health applications and returns to their stakeholders.
A
Findings help the government in identifying and communicating medical practitioners regarding
M
immunization or health-related programs so that they can plan for the medical training and visits
TE D
for improving treatment and care. It also assists the government in initiating and designing different insurance and health policy schemes for the benefit of the people particularly in the regions where access to health care is quite less.
EP
In future work, we intend to explore the perception and experience of the people in using mobile health applications such as LoseIt or mySugr Diabetes Logbook which help the application
CC
developers or service providers to provide better quality service. Moreover, causal loop diagram can be transformed to stock and flow diagram to measure the dynamics of different variables
A
influencing the accessibility and acceptability of the mobile health applications.
Author Contributions All authors have contributed to the writing of this research paper.
Acknowledgment This study did not receive any specific grant from funding agencies in the public, commercial, or
IP T
not-for-profit sectors. References
R. Ghosh, A. Datta, K. Lahiri, International OR Healthcare in emerging countries,
SC R
[1]
Challenges and opportunities for operations research, ORMS-Today (2013). [2]
Vital Wave Consulting, mHealth for Development: The Opportunity of Mobile Technology
first ed., Springer, New York, 2006.
A. Chib, M. H. van Velthoven, J. Car, mHealth adoption in low-resource environments: a
M
[4]
N
R. Istepanian, S. Laxminarayan, C. Pattichis, M-Health: emerging mobile health systems,
A
[3]
U
for Healthcare in the Developing World, 46 (2009) 1–70.
TE D
review of the use of mobile healthcare in developing countries, Journal of Health Communication, 20 (2015) 4-34. [5]
A. Gautam, Top Microblogging sites list with high pr and best 2016. My WP Tips.
[6]
EP
http://mywptips.com/top-microblogging-sites-list/ (accessed 3 May 2017). A. Agarwal, B. Xie, I. Vovsha, O. Rambow, R. Passonneau, Sentiment analysis of twitter
CC
data, In Proceedings of the Workshop on Languages in Social Media, Association for Computational Linguistics, USA, 2011, 30-38. P. Germanakos, C. Mourlas, G. Samaras, A mobile agent approach for ubiquitous and
A
[7]
personalized eHealth information systems. In Proceedings of the Workshop on ‘Personalization for e-Health ‘of the 10th International Conference on User Modeling (UM'05), Edinburgh, 2005, 67-70.
[8]
R. Evans, IMS Institute for Healthcare Informatics, Nursing Standard, 28 (2014) 30-30.
[9]
S. T. Liaw, Patient and general practitioner perceptions of patient-held health records, Family Practice, 10 (1993) 406-415.
[10] P.A. Abbott, J. Foster, H. de Fatima Marin, P.C. Dykes, P. C, Complexity and the science of
IP T
implementation in health IT—knowledge gaps and future visions, International Journal of Medical Informatics, 83 (2014) 12-22.
SC R
[11] Y. Jiang, S. M. Sereika, A. D. Dabbs, S. M. Handler, E. A. Schlenk, Using mobile health
technology to deliver decision support for self-monitoring after lung transplantation, International Journal of Medical Informatics, 94 (2016) 164-171.
U
[12] M. Siddiqui, M. Y. Islam, B. A. I. Mufti, N. Khan, M. S. Farooq, M. G. Muhammad, M.
N
Osama, D. Kherani, A. M. Kazi, Assessing acceptability of hypertensive/diabetic patients
A
towards mobile health based behavioral interventions in Pakistan: a pilot study, International
M
Journal of Medical Informatics, 84 (2015) 950-955.
TE D
[13] J. D. Sterman, Business Dynamics: systems thinking and modeling for a complex world, Irwin-McGraw-Hill, New York, 2000. [14] T. Binder, A. Vox, S. Belyazid, H. Haraldsson, M. Svensson, Developing system dynamics
EP
models from causal loop diagrams, In Proceedings of the 22nd International Conference of the System Dynamic Society, England, 2004, 1-21.
CC
[15] N. B. Lassen, R. Madsen, R. Vatrapu, Predicting iPhone sales from iPhone tweets. 2014 IEEE 18th International Enterprise Distributed Object Computing Conference (EDOC)
A
(2014) 81-90.
[16] P. Sedarati, A. Baktash, Smart glasses adoption in smart tourism destination: A conceptual model. In ENTER2017 Conference on Information and Communication Technologies in
Tourism. Rome, Italy. http://ertr.tamu.edu/content/issues/enter-2017-volume-8-researchnotes/ (accessed 13 June 2017) [17] S. Kanungo, Understanding the dynamics of information system use, In 21st International Conference of the System Dynamics Society, New York City, USA, 2003, 1-18.
IP T
[18] M. Mutingi, System dynamics of information technology adoption in a complex environment, International Journal of Industrial and Systems Engineering, 17 (2014) 78-96.
SC R
[19] D. Fisher, J. Norvell, S. Sonka, M. Nelson, Understanding technology adoption through
system dynamics modeling: implications for agribusiness management, The International Food And Agribusiness Management Review, 3 (2003) 281-296.
U
[20] Y. Chen, Understanding technology adoption through system dynamics approach: A case
N
study of RFID technology, 2011 IFIP 9th International Conference on Embedded and
A
Ubiquitous Computing (EUC), IEEE, Melbourne, Australia, 2011, 366-371.
M
[21] R. Wong, S. Y. Sheng, A system dynamics approach to the study of word-of-mouth and its
TE D
effect in an online environment, Technology Innovation Management Review, (2011) 4248.
[22] Y. Liu, H. Li, V. Kostakos, J. Goncalves, S. Hosio, F. Hu, An empirical investigation of
EP
mobile government adoption in rural China: A case study in Zhejiang province, Government Information Quarterly, 31 (2014) 432-442.
CC
[23] J. Cho, M. Quinlan, D. Park, G. Noh, Determinants of Adoption of Smartphone Health Apps among College Students, American Journal of Health Behavior, 38 (2014) 860-870.
A
[24] B. Liu, Sentiment analysis and subjectivity, In Handbook of Natural Language Processing, second ed., Chapman and Hall/CRC, 2010, 627-666.
[25] D. Cummings, H. Oh, N. Wang, Who needs polls? gauging public opinion from twitter data, 2010, Unpublished manuscript. [26] Johnson,
G.
Theories
of
Emotion
|
Internet
Encyclopedia
of
Philosophy.
http://www.iep.utm.edu/emotion/ (accessed 20 July 2017).
IP T
[27] A. Maslow, K. J. Lewis, Maslow's hierarchy of needs, Salenger Incorporated, 14 (1987).
[28] C. P. Alderfer, An empirical test of a new theory of human needs, Organizational Behavior
SC R
and Human Performance, 4 (1969) 142-175.
[29] E. A. Fagenson, Mentoring—Who needs it? A comparison of protégés' and nonprotégés' needs for power, achievement, affiliation, and autonomy, Journal of Vocational Behavior,
U
41 (1992) 48-60.
N
[30] J. Panksepp, Affective neuroscience: The foundations of human and animal emotions.
A
Oxford university press, 2004.
TE D
2nd ed., 2000, 91-115.
M
[31] L. Cosmides, J. Tooby, Evolutionary psychology and the emotions. Handbook of emotions,
[32] R. M. Nesse, Evolutionary explanations of emotions. Human nature, 1 (1990) 261-289. [33] R. Plutchik, Emotions: A general psychoevolutionary theory, Approaches to emotion,
EP
psychology press, 1984, 197-219.
[34] P. E. Griffiths, Is emotion a natural kind? In R. C. Solomon (Ed.), Thinking about feeling:
CC
Contemporary philosophers on emotions, Oxford university press, New York, 2004, 233– 249.
A
[35] C. Lutz, Unnatural emotions: Everyday sentiments on a Micronesian atoll & their challenge to western theory, Chicago: university of Chicago press, 1988.
[36] Harré, R, The social construction of emotions, Basil Blackwell, 1986.
[37] D. Centeno, Nurtured Emotional Evolution among the Ifaluk. Apathy and Prophecies. https://astrowipe.wordpress.com/2009/07/03/nutured-emotional-evolution-among-theifaluk/ 2009, (accessed 20 July 2017). [38] S. Rauch, E. Foa, Emotional Processing Theory (EPT) and Exposure Therapy for PTSD.
IP T
Journal of Contemporary Psychotherapy. 36 (2006) 61-65.
[39] M. Nussbaum, Emotions as judgements of value and importance. In R. C. Solomon (Ed.),
SC R
Thinking about feeling: Contemporary philosophers on emotions, Oxford university press, 2004, 183–199.
[40] I. J. Roseman, Cognitive determinants of emotion: A structural theory, Review of Personality
U
& Social Psychology, 1984.
N
[41] D. G. Myers, Myers' psychology for AP, Macmillan, 2010.
A
[42] V. S. Pagolu, K. N. R. Challa, G. Panda, B. Majhi, Sentiment Analysis of Twitter Data for
M
Predicting Stock Market Movements, 2016 International Conference on Signal Processing,
1345-1350.
TE D
Communication, Power and Embedded System (SCOPES), IEEE, Odisha, India, 2016,
[43] M. Rastegar-Mojarad, Z. Ye, D. Wall, N. Murali, S. Lin, Collecting and Analyzing Patient
EP
Experiences of Health Care From Social Media, JMIR Research Protocols, 4 (2015) e78. [44] C. Wong, M. Sap, A. Schwartz, R. Town, T. Baker, L. Ungar, R. Merchant, Twitter
CC
Sentiment Predicts Affordable Care Act Marketplace Enrollment, Journal of Medical Internet Research. 17 (2015) e5, 1 1-8.
A
[45] M. J. Paul, M. Dredze, You are what you Tweet: Analyzing Twitter for public health, Fifth International AAAI Conference on Weblogs and Social Media, AAAI Publications, 20 (2011) 265-272.
[46] I. Korkontzelos, A. Nikfarjam, M. Shardlow, A. Sarker, S. Ananiadou, G. H. Gonzalez, Analysis of the effect of sentiment analysis on extracting adverse drug reactions from tweets and forum posts, Journal of Biomedical Informatics. 62 (2016) 148-158. [47] M. Salathé, S. Khandelwal, Assessing vaccination sentiments with online social media:
IP T
implications for infectious disease dynamics and control, PLoS Computational Biology, 7 (2011), e1002199.
SC R
[48] F. D. Davis, R. P. Bagozzi, P. R. Warshaw, User acceptance of computer technology: a
TE D
M
A
N
U
comparison of two theoretical models. Management science, 35 (1989) 982-1003.
A
CC
EP
Figures
IP T SC R U N
A
CC
EP
TE D
M
A
Figure 1: Flow diagram for extracting tweets (single column fitting image)
Figure 2a: Bigram for fitness applications (single column fitting image)
IP T SC R U
800
741
A
Sentiment analysis of tweets about fitness app (classification by polarity)
M
600 500 400 300
TE D
Number of tweets
700
N
Figure 2b: Word cloud for fitness applications (2-column fitting image)
200 100 0
EP
Positive
202 57 Negative
Neutral
A
CC
Figure 3a: Sentiment analysis for fitness app based on polarity (single column fitting image)
800
742
Sentiment analysis of tweets about fitness app (classification by emotion)
700
500 400 300 200
132 71
100
16
16
Fear
Anger
Unknown
Joy
Sadness
16
SC R
0
IP T
Number of tweets
600
Surprise
EP
TE D
M
A
N
U
Figure 3b: Sentiment analysis for fitness app based on emotion (2-column fitting image)
A
CC
Figure 4a: Bigram for diabetes applications (single column fitting image)
7
Disgust
IP T SC R U
259
M
200 150
TE D
Number of tweets
250
100
EP
50 0
Sentiment analysis of tweets about diabetes app (classification by polarity)
A
300
N
Figure 4b: Word cloud for diabetes applications (2-column fitting image)
Positive
56
50
Negative
Neutral
A
CC
Figure 5a: Sentiment analysis for diabetes app based on polarity (single column fitting image)
350
Sentiment analysis of tweets about diabetes app (classification by emotion)
300
250 200
100 50
28
23
8
0 Unknown
Sadness
Joy
Anger
4
IP T
150
SC R
Number of tweets
300
Fear
2
Surprise
EP
TE D
M
A
N
U
Figure 5b: Sentiment analysis for diabetes app based on emotion (2-column fitting image)
A
CC
Figure 6a: Bigram for meditation applications (single column fitting image)
IP T SC R
900
810
Sentiment analysis of tweets about meditation app (classification by polarity)
N
800
U
Figure 6b: Word cloud for meditation applications (2-column fitting image)
A
600
M
500 400 300
TE D
Number of tweets
700
200 100 0
EP
Positive
112
Negative
78
Neutral
A
CC
Figure 7a: Sentiment analysis for meditation app based on polarity (single column fitting image)
1000 900
Sentiment analysis of tweets about meditation app (classification by emotion)
879
700 600 500 400 300 200 81 18
8
Disgust
Sadness
0 Unknown
Joy
8
SC R
100
IP T
Number of tweets
800
Surprise
6
Anger
EP
TE D
M
A
N
U
Figure 7b: Sentiment analysis for meditation app based on emotion (2- column fitting image)
A
CC
Figure 8a: Bigram for cancer application (single column fitting image)
IP T SC R
Sentiment analysis of tweets about cancer app (classification by polarity)
N
600
U
Figure 8b: Word cloud for cancer application (2-column fitting image)
491
A M
400
300
TE D
Number of tweets
500
200
EP
100
0
172
164
Positive
Negative
Neutral
A
CC
Figure 9a: Sentiment analysis for cancer app based on polarity (single column fitting image)
700
654
Sentiment analysis of tweets about cancer app (classification by emotion)
500 400
200 100
60
43
32
Sadness
Fear
0 Unknown
Joy
20
IP T
300
SC R
Number of tweets
600
Anger
TE D
M
A
N
U
Figure 9b: Sentiment analysis for cancer app based on emotion (2-column fitting image)
A
CC
EP
Figure 10a: Boxplot for mobile health applications (2-column fitting image)
18
Surprise
IP T SC R U
CC
EP
TE D
M
A
N
Figure 10b: Sentiment scores for mHealth applications (2-column fitting image)
Figure 11: Causal loop diagram showing the relationship between the potential adopters, semi-
A
adopters, and adopters of mHealth system
IP T SC R U N A
M
Figure 12: Causal loop diagram for perceived usefulness of mHealth system
TE D
Table 1: Total number of tweets extracted mHealth apps Tweets requested Tweets obtained 1000
1000
Diabetes app
1000
365
Meditation app
1000
1000
Cancer app
1000
827
CC
EP
Fitness app
Table 2: Tweets for fitness applications Tweets
Emotion
Polarity
A
Jenn at jensfunfinds says her favorite thing about japrasportpace is the Unknown Positive fitness test that comes with the app Officially logged my meals into my fitness pal app for days in a row Fear highly recommend it to everyone takes minutes
Positive
NL sports coming soon to a sports team near you sport coaching training Joy
Positive
apps tech fitness Can I put on my fitness app that I did in of squats because I squatted Anger
Negative
down to get my hidden treats and I was too high to get back up Sadness
Neutral
Discover why today is an amazing day to get started on that fitness or Surprise
Neutral
IP T
Get a free day pass to best fitness activate now with my special code
business goal
Genuinely paying pound a month for a fitness app if this won’t motivate Disgust
SC R
me nothing will be tired of feeling crap Table 3: Tweets for diabetes applications Tweets
Emotion
Neutral
Polarity
U
A mobile app for the self-management of type diabetes among Unknown Positive
N
adolescent a randomized controlled trial
Wear your medical alert bracelet or update the health on your Fear
A
smartphone, so others know you have type diabetes
or deal with diabetes very cool
M
Glad to hear I saw this Israeli app that can personalize diets to prevent Joy
Anger
TE D
He knows how to scan the QR code and get the diabetes risk score app
At a time when insulin amp other diabetes drugs are soaring Amazon Surprise
Positive
Positive
Negative Negative
creates a stupid distraction diabetes app contest duped again, folks Discover how breakthrough diet sensor nutrition app helps in managing Sadness
Neutral
EP
diabetes weight loss and improves fitness
CC
Table 4: Tweets for meditation applications Tweets
Emotion
Polarity
Meditation app headspace hires a new chief business officer tech crunch Unknown Positive Positive
A
Is the best app in fall asleep to the meditation sleep times in likes 25 Joy seconds amazing How to deal with pesky thoughts in meditation success
Anger
Headspace’s revamped app helps busy people turn meditation into habit Disgust wellness
Negative Negative
Heal your body mind and soul with these amazing meditation on iTunes Surprise
Neutral
for iPhone Wow just discovered meditation apps is there such a thing as a get off Sadness
Neutral
your butt amp write app cos hearing katching amwriting Table 5: Tweets about cancer applications Emotion
Polarity
IP T
Tweets
Sharktankaucancer touches us all so I really want this app to succeed Unknown Positive
SC R
amp make life impact changes Our nutrition app can help you find the optimal diet for any stage of your Fear cancer journey
Great to see innovative thinkers in the cancer care space sharktankau
Joy
U
Cancer app rides high on emotion surfer creates comic relief for patient Anger
N
brain tumor
Get the free mindfulness app for those living with cancer and their Sadness
A
caregivers
more than k people in countries
Positive Negative
Negative
Neutral
M
Cancer aid is the top cancer app in Australia UK amp the US used by Surprise
Positive
mHealth apps
TE D
Table 6: Boxplot values for mobile health apps median
Quartile values
Adjacent values
upper
lower
upper
lower
-1
-1
-
+1
-1
Diabetes
+1
-
+1
+2
-1
Meditation
0
+1
-
+2
-1
Cancer
0
+1
-
+2
-1
A
CC
EP
Fitness