=Paper=
{{Paper
|id=Vol-3147/paper11
|storemode=property
|title=Does behaviour match user typologies? An exploratory cluster analysis of behavioural data from a gamified fitness platform
|pdfUrl=https://ceur-ws.org/Vol-3147/paper11.pdf
|volume=Vol-3147
|authors=Jeanine Krath,Adam Palmquist,Izabella Jedel,Isak Barbopoulos,Miralem Helmefal,Robin Isfold Munkvold
|dblpUrl=https://dblp.org/rec/conf/gamifin/KrathPJBHM22
}}
==Does behaviour match user typologies? An exploratory cluster analysis of behavioural data from a gamified fitness platform==
https://ceur-ws.org/Vol-3147/paper11.pdf
Does behaviour match user typologies? An exploratory cluster
analysis of behavioural data from a gamified fitness platform
Jeanine Krath1, Adam Palmquist2, Izabella Jedel3, Isak Barbopoulos4, Miralem Helmefalk5 and
Robin Isfold Munkvold3
1
University of Koblenz-Landau, Universitaetsstrasse 1, Koblenz, 56070, Germany
2
University of Gothenburg, Forskningsgången 6, Gothenburg, 417 56, Sweden
3
Nord University, Universitetsalléen 11, Bodø, 8026, Norway
4
Insert Coin, Vasagatan 33, Gothenburg, 411 37, Sweden
5
Linnaeus University, Universitetsplatsen 1, Kalmar, 392 31, Sweden
Abstract
A promising solution to increase user engagement in gamified applications is tailored
gamification design. However, current personalisation relies primarily on user types identified
through self-reporting rather than actual behaviour. As a novel approach, the present study used
an exploratory machine learning analysis to identify seven clusters of users in a gamified fitness
application based on their behavioural data (N = 19,576). The clusters were then conceptually
compared to common user typologies in gamification, identifying possible relationships
between behavioural user clusters and user types motivated by achievement, sociability, and
extrinsic incentives. The findings shed light on nuanced behaviour patterns of user types in the
fitness context and how knowing these patterns can inform the way in which tailored
gamification could be implemented to meet the needs of specific types. Thereby, they contribute
to the discussion on utilising behavioural data and user typologies for tailored gamification
design.
Keywords 1
Cluster analysis, user types, tailored gamification design, personalisation, fitness, exploratory
machine learning, k-means clustering
1. Introduction individual differences in how gamification is
perceived and used to guide personalized
gamification design [5]. Instead of executing a
Gamification, the application of game
one-size-fits-all design, the prospect of
elements in a non-game context [1], has been
personalisation and adaptivity affords a design
researched within several fields to increase user
that can be automatically informed, rearranged,
engagement and motivation [2]. One of the most
and redeployed based on users’ actions and
popular applications of gamification is using
reactions [5]. Therefore, personalising gamified
game elements in fitness applications [3,4].
fitness applications might present a solution to
However, mixed outcomes have let gamification
create improved experiences and simultaneously
research both in general [5] and in the fitness
provide users with a wider array of features that
context [6,7] question the applicability of
may be of particular interest.
universal design and increasingly focus on
6th International GamiFIN Conference 2022 (GamiFIN 2022),
April 26-29, 2022, Finland
EMAIL: jkrath@uni-koblenz.de (A. 1); adam.palmquist@ait.gu.
se (A. 2); izabella.jedel@hotmail.com (A. 3); isak@insert
coin.se (A. 4); miralem.helmefalk@lnu.se (A. 5); robin.
munkvold@nord.no (A. 6)
ORCID: 0000-0003-4996-1147 (A. 1); 0000-0003-0943-6022 (A.
2); 0000-0001-9212-3259 (A. 3); 0000-0003-2485-0184 (A. 4);
0000-0003-2924-2874 (A. 5); 0000-0002-2524-279X (A. 6)
©️ 2022 Copyright for this paper by its authors. Use permitted under Creative
Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
105
Previous research on tailored gamification has 2. Background
explored personalised gamification design based
on demographic data such as age and gender, as 2.1. User typologies in gamification
well as personality [5,8]. However, to date, the
most widely used approach to personalising Personalisation of gamification design has
gamification interventions has been to classify recently gained tremendous importance in
users based on their needs and motivations via gamification research [19]. Within this stream,
user typologies [5]. A variety of user typologies researchers have proposed a variety of
have already been used in research on tailored classifications or typologies of users [5,8–10] and
gamification design [5,8–10], the most popular explored their preferences for game elements
among them being Bartle's player types [11], the [20,21] to inform tailored gamification design.
gamification user types HEXAD [12], the One of the first typologies was Bartle's
BrainHex typology [13] and Yee's motivations to typology of players in multi-user dungeons [11].
play MMORPGs [14]. Specifically, Yee's He distinguished between Achievers (who focus
motivations [15] and HEXAD types have been on earning points and levelling up), Explorers
successfully used to personalise gamification for (who enjoy discovering interesting features and
fitness and health applications [6,7]. exploring the system), Socializers (who value
A particular limitation of current gamification relationships with other players), and Killers (who
design based on these predefined user typologies like disrupting the experience of others) [11].
is that the design process mainly relies on Building on Bartle's findings, Yee [14] sought
questionnaires to determine user types based on to identify the underlying motivations of users of
self-report rather than actual behaviour, which is MMORPGs and found ten motivations
particularly challenging as user types can change categorised into the three components of
over time [16] and people tend to respond in self- Achievement (including advancement, mechanics
reports in a socially desirable way that does not and competition), Social (including socialising,
necessarily reflect their actual behaviour [17]. In relationships and teamwork) and Immersion
addition, users may be classified as multiple or (including discovery, role-playing, customisation
hybrid user types depending on the context [18]. and escapism) [14].
Therefore, further research is needed on how The BrainHex model by Nacke et al. aimed to
technologies such as machine learning [16] could combine findings from previous models with
help identify different user types based on their neurobiological insights [13]. As a result, they
behaviour and dynamically adapt the gamification described seven types of players, namely Seekers
system accordingly [16,18]. (who are curious and eager to explore), Survivors
The present study addresses this gap by (who enjoy intense fright experiences),
analysing user behaviour in a gamified fitness Daredevils (who enjoy thrilling and risky
application through a machine learning approach experiences), Masterminds (who like to overcome
and discussing the relationships between problems and develop strategies), Conquerors
identified clusters of users based on their actions (who derive satisfaction from defeating others),
and common user typologies. The contribution of Socializers (who like to interact with others) and
this work is thus twofold: Achievers (who are goal-oriented and motivated
First, a k-means cluster analysis is conducted by long-term success) [13].
to identify distinct clusters of users based on their While the previous typologies were developed
actions in a gamified fitness platform, drawing on in the context of games, Marczewski designed and
a large dataset (n = 19,576) of behavioural data developed the HEXAD typology, which builds on
with over 1 million events recorded in 49 weeks. extrinsic motivation and the three basic
Second, by exploring the extent to which the psychological needs [22] as a model specifically
identified clusters can be mapped to common user for use in gamification [12]. The model consists
typologies, we contribute to the ongoing of six user types: Philanthropists (driven by
discussion of user typologies in terms of tailored, purpose and altruism), Socialisers (driven by
personalised, and adaptive gamification. In social relations and interaction with others),
summary, the study aims to answer the following Achievers (striving for self-improvement through
research question: challenges and proficiency), Players (striving for
RQ1: Which clusters of users can be identified external rewards), Free Spirits (driven by
using exploratory clustering techniques on autonomy and exploration), and Disruptors
behaviour data from a gamified fitness platform?
106
Table 1
Prevalent user typologies in gamification research and their conceptual relations (based on [8,9,13])
Concept ([9]) Bartle ([11]) HEXAD ([12]) BrainHex ([13]) Yee ([14])
Achievement Achiever Achiever Achiever / Mastermind Achievement
Exploration Explorer Free Spirit Seeker Immersion
Sociability Socializer Socializer / Philanthropist Socializer Social
Domination Killer Disruptor Conqueror -
Gaming intensity and - - Survivor / Daredevil -
skill
- - Player - -
(driven by challenging the status quo and motivations in different age groups and found
participating in disruptive alterations) [12]. distinct preferences for intrinsic, extrinsic, and
All of these typologies share common feedback elements [27]. Recently, both Altmeyer
concepts that are reflected in certain types, such et al. [6] and Zhao et al. [7] used the HEXAD
as achievement, exploration or sociability [9], and typology to personalise gamified fitness apps and
several researchers have attempted to relate the found that it led to more positive affective
different user types to each other [8,9,13] (see experiences [6], motivation, and satisfaction [7]
Table 1 for an overview). For example, the than a one-size-fits-all approach.
concept of achievement is prevalent in each of the These previous approaches relied on self-
four typologies, while the HEXAD model adds report to obtain information about users'
the Player as a type that is extrinsically rather than motivations and needs, categorised into user types
intrinsically driven [6]. In contrast, the BrainHex and assumed to influence behaviour. However,
model describes Survivors and Daredevils further research is warranted on user types that
motivated by intense gaming experiences absent can be identified by analysing actual behaviour in
in other typologies [13]. gamified systems rather than relying on self-
reported motivations [16], especially since user
2.2. Personalisation in gamified types can manifest in hybrid or multiple forms
depending on the context [18].
fitness applications A promising approach to identifying different
types of users is to use unsupervised machine
Health and fitness is the second largest area of learning techniques [16] to cluster users based on
research in gamification [2]. Previous studies their recorded behaviours in the system. A recent
have shown that gamification in fitness tracking review has shown that automatically adapting
apps can successfully promote physical activity gamification design using machine learning is
and bodyweight reduction [23]. In the fitness gaining momentum [28]. It has been suggested
context, personalisation of gamification involves that artificial intelligence and machine learning
real-time adjustment of difficulty based on are among the most promising emerging areas in
physiological parameters such as heart rate [24] or gamification research [29]. Previous studies have
acceleration [25]. However, their usefulness for a applied clustering techniques in education to
thoroughly tailored gamification design that identify different types of students [30,31].
modifies different aspects of gamification with However, as far as the authors know, such
appropriate solutions for each user [26] is limited. techniques have not yet been applied to gamified
Therefore, other studies focusing on personalising fitness applications.
gamified fitness applications drew on
psychological determinants such as motivations 2.3. AUTOMATON project
and user typologies. One of the first studies on
individual differences in gamified fitness
applications was that of Brauner et al. [15], who The current work is based on a university and
observed that users' motivations to play [14] industry initiative started as an applied artificial
significantly influenced performance. Later, intelligence (A.I.) project between the University
Kappen et al. examined various exercise of Skövde and Insert Coin. The long-term project
107
goal is to design and develop a system of machine platform ecosystem, aiming for a gameful
learning models that are capable of independently ambience in the whole platform. In order to create
identifying user clusters and their behaviour this ambience, several motivational
patterns (as stage 1) and then make personalised affordances [2] were used, associated with
suggestions, as well as apply and/or adjust the different features in the platform ecosystem
gamification balance to better fit the users in each (Table 2), exemplified in Figure 1.
segment (as stage 2). The expected project
outcome is an adaptive gamified fitness platform 3.2. Procedure
based on predicted user preferences toward
tailored gamification experiences.
The platform has an event-based architecture,
meaning that user-generated events, such as
3. Method reaching training goals associated with milestones
3.1. Materials (e.g., 30 mins of activity, lifting a total of 3 tonnes
of weight in a workout), recording a new exercise
The cloud-based fitness platform in the study (e.g., power walking or weightlifting), viewing
is an iOS/Android application. The platform's planned activities or responding to a fitness coach
principal purpose is to function as a marketplace trial were logged for analysis. Therefore, the
between individuals interested in fitness and dataset for the cluster analysis consisted of
exercise on the one hand (hereinafter users) and 1,116,126 events recorded on the fitness platform
various fitness centres and fitness coaches on the originating from 19,576 unique users collected
other. The platform provides each user with a over 49 weeks.
workout diary to log various physical activities In order to cluster the user actions, the
(e.g., cycling, weightlifting, running), log and 1,116,126 events first had to be parsed (i.e.,
track their weight, and view other metrics variables and value labels had to be extracted from
indicating their overall progression. The platform the event meta-data) and aggregated into a list of
also includes social features, such as reactions, events associated with each cluster. As some
adding and sending messages to friends, posting event types consisted of both predefined activities
workouts for others to see, or browse, like or (e.g., "power walk") and free-text entries (e.g., "I
commenting on friends' workout feeds, as well as went running"), the total number of unique
connecting and interacting with other users or categorical values exceeded 1,600. Because of the
fitness coaches. large number of categorical values, we treated all
The gamification design was focused on the categorical values as text entries and used
workout diary due to its central position in the techniques suitable for this kind of data [31].
Figure 1: Screenshots of the fitness platform showing the overview of planned and logged activities,
a challenge with adaptive difficulty, the social feed with peer rating features and the personal profile
with points, virtual coins as currency for real-world rewards, level, milestones and statistics
108
Table 2 all clusters (TF-IDF) and descriptive statistics
Ecosystem's motivational affordances (e.g., size of the cluster relative to the whole
categorised after Koivisto & Hamari [2] sample, proportion of total events, and conversion
Affordance Category events, i.e., bought subscriptions, sent by users in
Experience points Progression the cluster).
Performance/ Achievement/ The clustering procedure and labelling were
Progress stats Progression planned and conducted by author 5, who was not
involved with the current research project at the
Milestones Achievement
time and therefore did not conduct the analysis or
Levels Achievement
set the labels with the purpose of relating them to
Team Leaderboards Achievement/Social
existing user typologies in gamified systems.
Challenges/Competitions Achievement/Social
Peer rating Social
Gameful narrative Immersion 4. Results
Adaptive difficulty Miscellaneous 4.1. Cluster model evaluation
Onboarding Miscellaneous
Real-world reward Miscellaneous
The model with the highest percentage of total
Reminders Miscellaneous
correct predictions across all its clusters was
Virtual currency Miscellaneous
selected as the final model. As a benchmark, the
most common event occurred on ~54% of the
First, each categorical value was split into target days. The "baseline" model (where all users
separate words. The words were then vectorised belonged to the same cluster) achieved a
with the Term Frequency–Inverse Document prediction score of 55%. However, the score
Frequency (TF-IDF) statistic2, a common way to increased to 68.5% in the cluster model with two
prepare text data for clustering in the field of clusters, and each subsequent model improved
natural language processing [32]. In order to this score further until a peak was reached in the
identify clusters, the vectorised events were then model with 7 clusters, achieving 76% correct
used as input data in a k-means cluster analysis predictions. Therefore, the model with 7 clusters
[33], conducted using the sci-kit-learn library in was selected.
Python [32]. The number of clusters (k) to extract
is often determined by the elbow method [33].
However, in this case, the ultimate aim of the 4.2. Identified clusters
analysis was to optimise the clusters specifically
for use in prediction models [34]. Therefore, a Once the final model had been selected, that
different approach was taken, in which each model was used to assign the entire sample of
cluster model (varying between 1 and 12 extracted 19,576 users to one of the seven identified
clusters) was evaluated in terms of how well the clusters. The clusters were then compared in terms
overall model could predict what event was most of most frequent events, most frequent events
likely to be recorded by users of a specific cluster offset by event frequency across all clusters (TF-
during their next active day (hereafter the target IDF), and various descriptive statistics (e.g., size
day), using a Sequential Long Short-Term and proportion of events sent by each
Memory model. Only users who had recorded cluster). Unless otherwise stated, all event
events on at least seven unique days were included frequencies are reported using the TF-IDF
in the training sample, which brought the sample statistic [32], not the absolute frequency, as this is
to a total of 9,667 users. Of these, 60% were used generally more useful to differentiate the clusters.
when training the models, and the remaining 40% Labels were set accordingly: Generalists
when testing the trained models and calculating (7.5% of users, 15.3% of events, 4.0% of
the prediction scores. The clusters were labelled conversions3), a cluster characterized by
based on their most frequently recorded events (in recording a relatively even spread between
absolute terms), their most frequently recorded physical activities (TF-IDF: power walk: 12.1%
events offset by the frequency of the event across of their total events, lift weights: 8.5%), scheduled
2 3
TF-IDF is especially useful for clustering, as the statistic will Triggered by paying for subscriptions.
increase proportionally to the number of times an event is recorded
by a user, offset by how many other users have recorded the same
event, thereby more effectively differentiating users.
109
activities (TF-IDF: perform planned activity: Table 3
19.8%, view planned activity: 5.0%, which can The clusters with their 5 most frequent events
both be done in the overview of planned and according to TF-IDF and their percentage of the
logged activities shown in Figure 1) and attaining clusters’ total events according to TF-IDF and
achievements (TF-IDF: active 30 minutes: 16.5%, absolute frequency (CF)
depicted on the personal profile in Figure 1); Cluster Top 5 events TF-IDF CF
Socializers (6.5% of users, 25.1% of events, 5.7%
Generalists Perform planned activity 19.8% 24.9%
of conversions), a cluster with a relatively high
n = 1,472 Achievement: 30 active min 4 16.5% 21.2%
degree of social activities (TF-IDF: like
(7,5%) Activity: Power walk 12.1% 15.1%
someone’s workout as depicted in the social feed
Activity: Lift weights 8.5% 10.5%
in Figure 1: 15.9% of their events; add friend:
View planned activities 5.0% 4.4%
3.4%); Achievement hunters (4.9% of users, 6.5%
of events, 2.6% of conversions), a cluster Socialisers Like someone's workout 15.9% 30.6%
characterized by a attaining a high degree of n = 1,279 Achievement: 30 active min. 7.8% 8.5%
(6,5%) Add friend 3.4% 2.0%
achievements during their workouts (TF-IDF: 30
Third-party application 5 3.2% 2.6%
active minutes: 18.2%, distance 3km: 8.4%);
Activity: Lift weights 2.7% 1.9%
Organizers (9.7% of users, 8.2% of events, 22.8%
Achievement Achievement: 30 active min. 18.2% 29.7%
of conversions), a cluster whose most frequent
hunters Activity: Walking 12.3% 18.4%
events involved viewing or planning scheduled
n = 960 Achievement: Distance 3km 8.4% 10.8%
activities (TF-IDF: view planned activity: 26.0%,
perform planned activity: 10.3%); Heavy lifters (4,9%) Activity: Lift weights 4.4% 5.5%
(14.8% of users, 6.7% of events, 21.9% of View planned activities 2.9% 2.7%
conversions), a cluster which prioritized Organizers View planned activities 26.0% 34.7%
weightlifting (TF-IDF: 23.4%) above most other n = 1,901 Perform planned activity 10.3% 13.1%
activities, and the only cluster where the (9,7%) Activity: Lift weights 7.9% 6.3%
achievement for lifting a total of 3000 kg during a Achievement: 30 active min. 7.1% 10.3%
workout appeared in the top 5; Weight watchers Activity: Running 4.8% 3.4%
(the most populous cluster by far, at 47.5% of Heavy lifters Activity: Lift weights 23.4% 29.9%
users, 20.6% of events, 37.1% of conversions), a n = 2,894 Perform planned activity 20.5% 22.8%
cluster where the users focus on tracking their (14,8%) Achievement: 30 active min. 13.7% 20.4%
weight progress more frequently than other users View planned activities 5.9% 5.3%
(TF-IDF: 12.8%); and finally Third-party app. Achievement: Lift 3000kg 3.4% 3.4%
total
users (9.1% of users, 17.5% of events, 5.9% of
Weight View weight progress 12.8% 2.4%
conversions), denote a cluster characterized by a
watchers Achievement: 30 active min. 9.2% 26.3%
high frequency of events related to third-party
n = 9,288 Third-party application 5.1% 1.2%
devices and apps (41.9% of total events fall in this
(47,5%) Activity: Lift weights 3.7% 5.5%
category according to TF-IDF). The clusters and
Achievement: Distance 3km 3.4% 7.7%
their most frequent events are shown in Table 3.
Third-party Third-party application 41.9% 60.9%
app. users Achievement: 30 active min. 5.8% 7.5%
5. Discussion n = 1,782 Activity: Lift weights 3.8% 3.6%
(9,1%) View planned activities 3.7% 2.7%
The results of our exploratory cluster analysis Perform planned activity 3.4% 3.4%
led to the identification of seven distinct clusters
of gamified fitness platform users based on their systems rather than actual behaviour [16,18]. In
behaviours. By applying k-means clustering, a order to elaborate on the contribution of this study
machine learning technique, to identify these to the scientific debate on personalisation of
clusters based on over one million events recorded gamified systems, which is currently oriented
by 19,576 users, the results extend previous towards needs- and motivation-based user
research [6,7,15,27] that mainly relied on self- typologies [5,8–10], it is important to discuss how
report tools as a basis for personalising gamified the exploratively identified clusters conceptually
4
“Achievement” events are generated when a user attains one of the etc.). We do not have any information about which apps users
achievements in the gamification design. interacted with or how they used them.
5
The “Third-party application” event refers to events generated by
other (third-party) devices/apps (like smart watches, smart scales,
110
relate to these typologies, in order to examine how constellations of behaviour related to these
certain needs can manifest themselves in achievement needs. Comparing the top five events
behaviours and how behaviour-based clusters of Organizers' and Heavy lifters shows that they
might contribute to effective tailored gamification record similar events, but with different
design, given the high predictability of future user frequencies. The most common events of
actions based on the identified clusters. The Organisers are View planned activity (26.0%)
conceptual discussion (see Table 4 for an and Perform planned activity (10.3%), which
overview) is based on the most distinguishing indicates a desire to plan and track activities and
events of each cluster (Table 3) and the described progress towards goals, reflecting the long-term
characteristics of the user types in the typologies. orientation postulated in the BrainHex typology
From a behavioural perspective, the Socialiser [13]. In turn, the most frequent event of the Heavy
cluster differs from the others in the prevalence of lifters is Lift weights (23.4%), followed by events
social events. The most common event is Like relating to planned activities, and it was the only
someone's workout (16.7%), probably to cluster with the achievement Lift 3000kg among
encourage others after a workout and show social their top 5 events (3.4%), suggesting that they
appreciation for their performance. User types might be motivated by the challenge of strenuous
driven by sociability [9], i.e., Socializers in physical activities and self-improvement through
Bartle's typology [11], HEXAD [12] and planning and mastery, which fits with the
BrainHex [13] and Social motivation in Yee's achievement-oriented user types reflected in the
motivations [14], are motivated by relatedness, HEXAD typology [12] and Yee's motivations
social connections and interaction [12] and like [14]. In comparison to Organisers, Heavy
social networks and social status functions lifters represent a more action-oriented cluster, as
[20,21]. Thus, we argue for a first conceptual they seem to be focused on mastering a specific
relationship between the Socialiser cluster and form of physical activity (weightlifting). At the
these socially-driven user types. same time, the former performs a more diverse set
Achievement hunters show a comparatively of physical activities and is more characterised by
dominant number of events related to attaining in- the planning itself. Generalists are distinguished
app achievements. They over proportionally by a more even spread of events between planning
earned achievements for 30 active minutes (Perform planned activity: 19.8%, View planned
(18.2%) and 3km distance (8.4%), with the most activities: 5.0%), attaining achievements (30
common activity being walking (12.3%), which active minutes: 16.5%) and performing physical
might be ideal to get these achievements. Players, activities (Power walk: 12.1%, Lift weights:
described as users who are motivated by extrinsic 8.5%). Like Organisers and Heavy lifters, they
affirmations of their achievements [12], are keen also seem motivated by goal-setting and challenge
on receiving virtual or real-world rewards and but less focused on either aspect. In contrast to
incentives for their activities [20,21]. Therefore, these more action- and planning-oriented clusters,
another link could be seen between the Generalists seem to combine the short-term
Achievement hunters cluster and the challenge and advancement orientation [12,14]
HEXAD Player, whereby we can only relate to with the long-term goal orientation [13].
the HEXAD because extrinsic motivations are not The Weight watchers cluster is fascinating
reflected in other typologies [9,11,12]. It should because it is the most populous cluster (47.5% of
be noted, however, that the causal relationship users) and, at the same time, is more difficult to
cannot be clearly determined (i.e., it could also be relate to existing user typologies. While it could
that they got to the achievements because they be argued that monitoring progress (View weight
mostly preferred walking rather than vice versa). progress: 12.8% of events) is related to an
Next, we see three clusters of users that best achievement orientation [14], the cluster lacks
relate to the concept of achievement through self- events indicating goal-oriented planning [13],
improvement [9]. Associated types are described which argues against a link to existing user types.
as motivated by levelling up [11], overcoming The Third party app. users cluster is
challenges [12], advancing and competition characterised by events that have been recorded
orientation [14] and goal orientation [13] and thus via appliances such as smartwatches and smart
report liking features such as progress monitoring scales. However, as we lack information on what
and levels as well as challenges [20,21]. In the events were recorded in them, we cannot conclude
clusters of Organisers, Heavy lifters, and the specific behaviours of this cluster and thus
Generalists, we can observe different cannot relate them to user types.
111
Table 4
Clusters postulated to be conceptually related to different user types in gamification research
Identified Cluster Bartle ([11]) HEXAD ([12]) BrainHex ([13]) Yee ([14])
Organizers / Achiever Achiever Achiever / Mastermind Achievement
Heavy lifters / Generalists
- Explorer Free Spirit Seeker Immersion
Socializers Socializer Socializer / Socializer Social
Philanthropist
- Killer Disruptor Conqueror -
- - - Survivor / Daredevil -
Achievement hunters - Player - -
Weight watchers - - - -
Third party app. users - - - -
There are other types from existing user simpler models with fewer clusters,
typologies, namely those driven by exploration personalisation based on these different kinds of
[9] (Explorer, Free Spirit, Seeker, Immersion), achievement-oriented behaviours may be more
and domination [9] (Killer, Disruptor, effective than one that groups them under a single
Conqueror), Philanthropists [12] and type. This insight is fascinating and calls for
Survivors/Daredevils [13] that could not be further research into the possible multi-
identified in the clusters. This is likely because the dimensionality of user types.
gamified fitness platform does not offer specific Second, we illustrate the value of behavioural
features corresponding to these user types, so no data for personalised gamification design. The
behavioural clusters emerged concerning these Weight watchers could not be clearly linked to
needs. For Philanthropists, there was no specific existing user typologies, yet it is the most
altruistic action [12] that users could perform, populous cluster and exhibits distinct behavioural
apart from liking others’ workouts, which we patterns from other clusters. This result does not
deem to be more related to the social aspect. allow conclusions to be drawn about general user
Furthermore, there was no specific way to express typologies, as the cluster probably results from a
autonomy and exploration [11–14], nor to particular type of monitoring that is likely to be
dominate other players [13] or disrupt their relevant only to fitness apps and similar platforms
experience [11,12]. Concerning Survivors and rather than a more general motivation or need.
Daredevils, the gamified fitness platform as a However, it does illustrate the value of basing
smartphone app might not have provided intense personalisation not only on user types but also on
and thrilling experiences. actual behaviours. For example, progress statistics
The preceding discussion yields interesting and milestones may generally appeal to users with
contributions to the debate on user typologies and a need for achievement [2], while for specific
behavioural data for personalised gamification users, as the Weight watchers cluster, progress
design. First, several distinct clusters of users and milestones may be primarily of interest if they
driven by achievement could be identified, likely related specifically to their weight. Likewise,
due to the fitness platforms nature and possibly users in the Heavy lifters cluster may prefer them
enabled by the wide variety of motivational related to their weightlifting goals. Thus, by
affordances in the Achievement and Progression clustering users based on behavioural data, the
category (Table 2). In examining the clusters, we emergent behavioural patterns can inform how a
observed a general, action-oriented, and planning- given gamification mechanic should be
oriented expression of achievement motivation implemented to meet their psychological needs.
types, suggesting that the actual behaviour
patterns of these user types may be more nuanced 6. Limitations and outlook
than theory would suggest. Furthermore, since the
cluster model that distinguished these three
The cluster of Third-party app. users was
achievement-oriented clusters outperformed
characterised by a high frequency of events
112
recorded via third-party applications (41.9%). 45 (2019) 191–210.
Thus, a limitation of the present research is that doi:10.0.3.248/j.ijinfomgt.2018.10.013.
we lacked information about what those [3] Y. Wang, W.B. Collins, Systematic
applications were and how they were used, so it is evaluation of mobile fitness apps: Apps as
difficult to accurately describe this cluster and the Tutor, Recorder, Game Companion, and
discuss to what extent it may relate to other Cheerleader, Telemat. Informatics. 59 (2021)
clusters and user typologies. Another limitation of 101552. doi:10.1016/j.tele.2020.101552.
the study is that we did not have self-reported data [4] J. Koivisto, J. Hamari, Gamification of
from questionnaires on user typologies, so the physical activity: A systematic literature
identified clusters could not be directly correlated, review of comparison studies, in: Proc. 3rd
but only conceptually related based on their most Int. GamiFIN Conf., 2019: pp. 106–117.
characteristic behaviours (as calculated using the [5] A.C.T. Klock, I. Gasparini, M.S. Pimenta, J.
TF-IDF statistic), which are merely hypotheses to Hamari, Tailored gamification: A review of
be explored in further studies. Therefore, future literature, Int. J. Hum. Comput. Stud. 144
research is encouraged to extend the behavioural (2020) 102495.
cluster analysis with data on user types and doi:10.1016/j.ijhcs.2020.102495.
correlate the self-reported responses with [6] M. Altmeyer, P. Lessel, S. Jantwal, L.
observed behaviours to gain a more nuanced and Muller, F. Daiber, A. Krüger, Potential and
comprehensive understanding of the relationship effects of personalizing gameful fitness
between needs-based user typologies and actual applications using behavior change
behaviour. As the gamification design of the intentions and Hexad user types, User Model.
application was not a priori based on matching User-Adapt. Interact. 31 (2021) 675–712.
gamification mechanics with different doi:10.1007/s11257-021-09288-6.
psychological needs, the results of the behavioural [7] Z. Zhao, A. Arya, R. Orji, G. Chan, Effects
cluster analysis, as well as the relationship to of a personalized fitness recommender
needs-based user typologies, may be different for system using gamification and continuous
other applications with different gamification player modeling: System design and long-
features or other target groups and domains (e.g., term validation study, JMIR Serious Games.
gamification in sustainability or education). For 8 (2020) 1–27. doi:10.2196/19968.
example, the gamified fitness platform used in this [8] F. De Vette, M. Tabak, M. Dekker - van
study did not include features related to Weering, M. Vollenbroek-Hutten, Engaging
exploration and domination, which is likely the Elderly People in Telemedicine Through
reason why the identified behavioural clusters Gamification, JMIR Serious Games. 3
could not be related to user types associated with (2015) e9. doi:10.2196/games.4561.
these motivations. In order to understand the [9] J. Hamari, J. Tuunanen, Player Types: A
generalisability of the results of our cluster Meta-synthesis, Trans. Digit. Games Res.
analysis, further research should be conducted Assoc. 1 (2014) 29–53.
using similar methods in gamified applications in doi:10.26503/todigra.v1i2.13.
different contexts. Finally, since need-based user [10] A. Mora, D. Riera, C. González, J. Arnedo-
types have been shown to change over time [16], Moreno, Gamification: a systematic review
it would be interesting to further study the of design frameworks, J. Comput. High.
behavioural clusters' temporal context and explore Educ. 29 (2017) 516–548.
whether they are stable over time or transient. doi:10.1007/s12528-017-9150-4.
[11] R. Bartle, Hearts, clubs, diamonds, spades:
7. References Players who suit MUDs, J. MUD Res. 1
(1996) 19.
[12] A. Marczewski, User Types, in: Even Ninja
[1] S. Deterding, D. Dixon, R. Khaled, L.E. Monkeys Like to Play, CreateSpace
Nacke, Gamification: Toward a Definition, Independent Publishing, 2015: pp. 65–80.
in: CHI 2011, 2010: pp. 1–4. [13] L.E. Nacke, C. Bateman, R.L. Mandryk,
doi:10.1007/978-3-642-13959-8_1. BrainHex: A neurobiological gamer
[2] J. Koivisto, J. Hamari, The rise of
typology survey, Entertain. Comput. 5
motivational information systems: A review (2014) 55–62.
of gamification research., Int. J. Inf. Manage. doi:10.1016/j.entcom.2013.06.002.
113
[14] N. Yee, Motivations for Play in Online Engagement among Adolescents? Results
Games, CyberPsychology Behav. 9 (2006) from A Cluster-Randomized Controlled
772–775. doi:10.1089/cpb.2006.9.772. Trial, Int. J. Environ. Res. Public Health. 18
[15] P. Brauner, A. Calero Valdez, U. Schroeder, (2021) 7444. doi:10.3390/ijerph18147444.
M. Ziefle, Increase Physical Fitness and [25] S.A. Pérez, A.M. Díaz, D.M. López,
Create Health Awareness through Personalized Tracking of Physical Activity
Exergames and Gamification, in: Hum. in Children Using a Wearable Heart Rate
Factors Comput. Informatics, 2013: pp. 349– Monitor, Int. J. Environ. Res. Public Health.
362. doi:10.1007/978-3-642-39062-3_22. 17 (2020) 5895.
[16] A.C.G. Santos, W. Oliveira, J. Hamari, S. doi:10.3390/ijerph17165895.
Isotani, Do people’s user types change over [26] V.M. García-Barrios, F. Mödritscher, C.
time? An exploratory study, in: Proc. 5th Int. Gütl, Personalisation versus Adaptation? A
GamiFIN Conf., 2021: pp. 90–99. User-centred Model Approach and its
[17] D.M. Randall, M.F. Fernandes, The social Application, in: Proc. Int. Conf. Knowl.
desirability response bias in ethics research, Manag., 2005: pp. 120–127.
J. Bus. Ethics. 10 (1991) 805–817. doi:10.1.1.161.7376.
doi:10.1007/BF00383696. [27] D.L. Kappen, P. Mirza-Babaei, L.E. Nacke,
[18] S. Hallifax, A. Serna, J.-C. Marty, G. Gamification through the application of
Lavoué, E. Lavoué, Factors to Consider for motivational affordances for physical
Tailored Gamification, in: Proc. Annu. activity technology, in: Proc. Annu. Symp.
Symp. Comput. Interact. Play, ACM, New Comput. Interact. Play, 2017: pp. 5–18.
York, NY, USA, 2019: pp. 559–572. doi:10.1145/3116595.3116604.
doi:10.1145/3311350.3347167. [28] A. Khakpour, R. Colomo-Palacios,
[19] S. Schöbel, M. Schmidt-Kraepelin, A. Convergence of Gamification and Machine
Janson, A. Sunyaev, Adaptive and Learning: A Systematic Literature Review,
Personalized Gamification Designs: Call for Technol. Knowl. Learn. 26 (2021) 597–636.
Action and Future Research, AIS Trans. doi:10.1007/s10758-020-09456-4.
Human-Computer Interact. 13 (2021) 479– [29] A. Mazarakis, Gamification Reloaded, I-
494. doi:10.17705/1thci.00158. Com. 20 (2021) 279–294. doi:10.1515/icom-
[20] J. Krath, H.F.O. von Korflesch, Player Types 2021-0025.
and Game Element Preferences: [30] D. Codish, E. Rabin, G. Ravid, User behavior
Investigating the Relationship with the pattern detection in unstructured processes –
Gamification User Types HEXAD Scale, in: a learning management system case study,
HCI Int. 2021, 2021: pp. 219–238. Interact. Learn. Environ. 27 (2019) 699–725.
doi:10.1007/978-3-030-77277-2_18. doi:10.1080/10494820.2019.1610456.
[21] G.F. Tondello, A. Mora, L.E. Nacke, [31] A.H. Nabizadeh, J. Jorge, S. Gama, D.
Elements of Gameful Design Emerging from Goncalves, How Do Students Behave in a
User Preferences, in: Proc. Annu. Symp. Gamified Course?—A Ten-Year Study,
Comput. Interact. Play, ACM, New York, IEEE Access. 9 (2021) 81008–81031.
NY, USA, 2017: pp. 129–142. doi:10.1109/ACCESS.2021.3083238.
doi:10.1145/3116595.3116627. [32] G. Varoquaux, L. Buitinck, G. Louppe, O.
[22] R.M. Ryan, E.L. Deci, Intrinsic and Extrinsic Grisel, F. Pedregosa, A. Mueller, Scikit-
Motivations: Classic Definitions and New learn, GetMobile Mob. Comput. Commun.
Directions, Contemp. Educ. Psychol. 25 19 (2015) 29–33.
(2000) 54–67. doi:10.1006/ceps.1999.1020. doi:10.1145/2786984.2786995.
[23] C. González-González, N.G. Río, V. [33] B. Purnima, K. Arvind, EBK-means: A
Navarro-Adelantado, Exploring the Benefits clustering technique based on elbow method
of Using Gamification and Videogames for and k-means in WSN, Int. J. Comput. Appl.
Physical Exercise: a Review of State of Art, 105 (2014) 17–24.
Int. J. Interact. Multimed. Artif. Intell. 5 [34] W. Jenq-Haur, L. Ting-Wei, L. Xiong, W.
(2018) 46. doi:10.9781/ijimai.2018.03.005. Long, An LSTM approach to short text
[24] A. Schwarz, G. Cardon, S. Chastin, J. sentiment classification with word
Stragier, L. De Marez, A. DeSmet, Does embeddings, in: 2018 Conf. Comput.
Dynamic Tailoring of A Narrative-Driven Linguist. Speech Process, 2018: pp. 214–
Exergame Result in Higher User 223.
114