Gamification personalization has been increasingly investigated as an avenue to improve the effects of gamification. While currently, empirical data exist to start making evidence-based gamification design, current guidelines and methods to bridge the gap of evidence and design is lacking. To start facing this challenge, we outline a point of departure proposing the recommendation system ReGammend (recommendation system for gamification designs). The system tailor gamification design based on users' traits, contextual factors, goals, and other relevant moderating factors. The recommendation system uses information from the previous literature to recommend gamification designs with multiple game elements aiming to positively affect the positive outcomes stemming from gamification. The proposed system contributes to researchers and practitioners, providing a practical way to personalize gamification designs.
In the last decade, gamification (i.e., the design approach of products, activities, services and systems to create similar motivational experiences as games usually create [1]), has increasingly become an important research topic in different contexts [2]–[5]. The application of gamification in contexts such as education [6], health [7], and government services [8], in general, seeks to affect the user behavior, engaging them during the use of gamified environments [9].
Different studies indicated that applying
gamification could have positive results in the
users, as students having better learning outcomes
[
Nowadays player and user typologies are the most investigated users’ characteristic in personalized gamification, with indications that the user preference over gamification designs depends on their user types [15]. Prior research has also indicated that the user types are dynamic [16], what would demand from designers and researchers a constant personalization of the gamified systems based on these user types. This process would be easier with the automation of the personalization, however, albeit the considerable number of studies that sought to personalize gamification over the years, the automation of this process still remains a lack in the field [3].
One prominent possibility to make the process of personalization easier for researchers and designers could be the use of recommendation systems (RS). RS uses prior information to create a more suitable suggestion for the users, and have been used especially in the e-commerce and entertainment industry [17]. The recommendations provided by a RS can be done based on several aspects, such as user demographic information or purchases historic [18] and would help the user to find what best suits its preferences between all the items available [19].
Thus, in the field of personalized gamification, RS can be a useful tool to recommend personalized designs, since they can indicate to users the gamified activities that would better fit to their preferences [17]. Thus, automation of personalized gamification with RS also could help designers implement gamification to users who have no previous experience in the usage of gamified systems, create a more efficient personalization, as well as avoid asking the user about their preferences constantly.
Albeit some studies have started to seek how to implement RS in the gamification context [17], [20], proposals of how to implement RS to define the gamification design remains a lack in the field. To start to face this challenge, in this paper we present a novel approach that is an evidence-based RS that provide recommendations of which would be the most suitable gamification design for each user type, according to the users’ traits. The RS proposed in this paper can be adapted and plugged-in different kind of gamified systems, thus, allowing designers and researchers to provide automatic recommendations for gamification design. At the same time, our work generates insights for future studies about dynamic recommendation of gamification designs in terms of graphical user interface (GUI).
In this section, we present the main topics
addressed in this paper (i.e., personalized
gamification and RS in gamification), and the
main related works. Personalization of
gamification has shown through recent research
that people have different orientations and
preferences regarding gamification design [15],
and therefore are affected differently according to
the type of gamification design they need to use
[
Albeit the different users’ characteristics that have been investigated, the player and user typologies have received major attention [3]. Over the years, researchers have worked on how different patterns could be grouped and therefore indicate different player/user types in games and gamification systems. These player/user types normally are grounded in motivational or psychological theories [27], player experiences [28], or even neurobiological research [29]. The choice of the player/user typology that would be used in a personalized gamified system, can be one major factor on the user motivation [30], and therefore, should be an important aspect to be considered in the development of gamified personalized systems.
One way that has recently started to be discussed to improve the personalization of gamification is the use of RS [17], which in short are systems/algorithms capable of identifying aspects of individual users and provide dynamic recommendations (e.g., design recommendations) [31]. RS can be classified into different categories: i) personalized, ii) collaborative, iii) content-based, iv) knowledge-based, and v) evidence-based [32].
The personalized recommendation systems use user profiles and some contextual parameters of users to provide personalized recommendations [32]. The collaborative recommendation systems use user-profile, some contextual parameters, and data of the community to which the user belongs. It recommends a similar product to a user which other users of their community are buying [32]. The content-based recommendation systems use the user-profile, contextual parameters as well as features of the product. Based on this, it recommends the product to the user which has the same feature as the product he has already purchased before [32]. The knowledge-based recommendation systems use the user profile, contextual parameters, product features, and knowledge models which keeps track of certain event in users’ demographics and accordingly do the recommendations (e.g., birthday recommends a certain product) [32]. Evidence-based recommendation systems use the user profile and previews evidence collected (e.g., results of research) [32]. Evidence-based recommendation systems were of particular interest to us as it allowed us to use previous acknowledgement from the literature to provide recommendations.
Over the years, different approaches have been used to provide dynamic adaptation of GUI in different areas [32]–[34]. In the field of gamification, some studies involving recommendation also have been conducted. Khoshkangini et al. [20] designed and implemented a fully automated system for the dynamic generation and recommendation of challenges, which are personalized and contextualized based on the preferences, history game status, and performances of each player. They conducted a long-running open-field experiment (12 weeks) involving more than 400 active participants, however, they focused on proposing recommendations for challenges in gamified systems without proposing recommendations for gamification design itself.
Herpich et al. [35] proposed a digital picture frame that interleaves a picture display mode with a recommender mode to promote a healthy lifestyle and to increase well-being of elderly people. Although they used gamification as a means to increase user appreciation of the system, the authors also did not provide recommendations directly related to gamification designs.
Su et al. [36] proposed an adaptative path RS for the teaching of geometry. The authors also proposed and evaluated a gamified prototype within the system. The results indicated that personalized recommendations are important [36], however, the authors also did not provide recommendations related to gamification design.
Tondello et al. [17] proposed a general framework for personalized gameful applications using RS (i.e., a framework to design RS for gamified applications). The framework proposed by Tondello et al. [17] does not provide a RS per se, but it helps the community to create RS for gamified systems.
Santos et al. [15] investigated how Hexad user types (i.e., Achiever, Disruptor, Free Spirit, Philanthropist, Player, and Socialiser) are associated with the preference and perceived sense of accomplishment from different gamification designs (Performance, Ecological, social, Personal, and Fictional). The study conducted by Santos et al. [15] provides insights into which gamification designs are suitable for each user type, however, does not provide practical approach to implement this personalization in gamified systems.
In summary, studies on the recommendation in gamified systems focus on personalizing system attributes (e.g., challenges and tasks), however, do not focus on personalizing the gamification design, and at the same time do not present how to automate the personalization process. At the best of our knowledge, this is the first evidencebased RS for gamification design. Table 1 present a comparison between the related work.
In this section, we present the RS proposed in this work. The system aims to provide recommendations of gamification designs according to the user’s types. In summary, the RS receives as input the user type and provide as output, a recommendation of gamification design for the user. An example of implementation is presented.
To design the general architecture of the RS, we used the framework proposed by Tondello et al. [17]. This framework defines the general inputs, processes, and outputs to implement recommendation in gamified applications. Tondello’s framework was of our interest because, as far we know, is the only framework to implement RS in gamified systems and can be adapted for different contexts.
To define/identify the users’ types, in the example of implementation presented in our paper, we used the Hexad framework [27], which defines different orientations of users according to their preferences related to interaction with gameful applications. The Hexad framework defines six different user types that a given user can be (i.e., Achiever, Disruptor, Free Spirit, Philanthropist, Player, and Socialiser). Hexad was of special interest in our work because it is (as far as we know) the only model for user types identification for the gamification domain. At the same time, it has already been validated in several languages and is widely used in academia and industry [37]–[39]. However, in future uses of our RS, Hexad can be replaced by another framework that better adapts to the application context.
To define the gamification designs to be recommended, in our example if implementation, we used the taxonomy proposed by Toda et al. [40]. Toda’s taxonomy defines five gamification designs that are organized according to motivation types and can be used to personalize gamified environments. The taxonomy proposed by Toda et al. [40] was especially used in our work because it is, as far as we know, the only taxonomy focused on proposing gamification designs, as well as because it is already widely used in field studies. Also, in future uses of our RS, the taxonomy can be replaced by another that better fits the application context.
Finally, to provide recommendations for gamification designs, in our example of implementation, we followed an evidence-based recommendation model, using the study conducted by Santos et al. [15], who identified how the different gamification designs proposed in the taxonomy of Toda et al. [40] affect the perceived sense of accomplishment and preference of users according to their Hexad user type. We used the study conducted by Santos et al. [15] as a basis for our recommendations for being, as far as we know, the only study that related Hexad user types with the gamification designs proposed by Toda et al. [40].
The work was organized in two general steps: i) RS design (general architecture) and ii) RS implementation (example of implementation). In the first step, the general idea of the RS was planned according to the materials previously described. In the second step, an example of implementation was provided, so that it could be used in different types of gamified systems. 3.2. design
Initially, the general RS was modeled according to Tondello’s framework [17]. The framework defines that a RS for gamification should have four Inputs (User profile, Items, Transactions, and Context), a Recommendation model, and a Rating [17].
The user profile should represent the user information that will be taken into account during the personalization process [17]. In our example, we used the Hexad profile of users [37]. Items must represent the system attributes that were used in the personalization process [17]. In our implementation example, we used the gamification design types proposed by Toda et al. [40].
The Transactions must represent how the personalization will be defined [17]. In our implementation example, transactions are the crossover between user types (Hexad) and gamification designs, defined according to the model by Santos et al. [15]. The Context must represent the definitions made at the user level of the system [17]. In our implementation example, we use information from the system administrator, who can make settings related to the type of personalization they want (accomplishment-based recommendation or preference-based recommendation).
The Recommendation must be the algorithm itself, where personalization is processed [17]. In our implementation example, we used an evidence-based algorithm to provide the gamification design recommendation according to the results of the study by Santos et al. [15]. Finally, Ratings are the recommendations generated by the algorithm [17]. In our implementation example, ratings are the gamification design recommendations that should appear in the user interface. Despite the examples used/suggested in this study (e.g., Hexad [27], Toda’s framework [40], and Santos’s recommendations [15]), the proposed RS is independent of these examples and can be adapted according to different needs. Figure 1 presents the general structure for our RS.
Finally, the RS was defined in nine components: i) User, ii) Admin, iii) User profile, iv) User modeling unit, v) User model, vi) System database, vii) Admin settings, viii) Content managing unit, and ix) User object (UO): User: the user is the person who will use the gamification system. The system can identify the trait, for instance, the user can answer a questionnaire (Hexad in our example) to provide their user type to the system (as input) and will receive the personalized system with the most appropriate gamification designs for their profile (as output). • Admin: The admin is responsible for managing the gamification system. It chooses which parameter to take into account when recommending a design. In the example of implementation provided in our work, the admin may choose between “user preference” or “user perceived sense of accomplishment” to the algorithm provide the recommendation. • User profile: The user profile consists of the answers to the questionnaire. • User modeling unit: The user modeling unit is the unit responsible for processing the user’s answers provided in the questionnaire. It returns the scores for each user type (e.g., Disruptor, Free Spirit, Achiever, Player, Socialiser, and Philanthropist) when using Hexad. • User model: The user model is responsible to stores the information returned by the User modeling unit. Therefore, it contains the scores of each user type. • System database: The system database contains data about previous study’ users so that the collaborative RS can compare them to the current user. Also contains the results obtained from the previous study and the variety of possible gamification designs. • Admin settings: The admin settings store information about which parameter the admin wants to use for recommending gamification designs. • Content managing unit: The content managing unit manages all the information coming from the User Model, System database, and Admin settings. It processes data to provide a rating for each possible gamification design. Returns the best design rating for the UO. • User object (UO): Contains the recommended gamification design for the current user.
All of the RS components can be changed as per system needs. In other words, where we use Hexad as a framework to identify user types, another framework that is more appropriate for each context type can be used (e.g., BrainHex [29], Bartle’s Archetypes [28]). Where we are using Toda’s taxonomy to define gamification designs [40], other more context-appropriate taxonomies can be used. Where we are using the study by Santos et al. [15] to define transactions, other evidence-based information can be used.
Figure 2 present the general RS architecture.
To provide a RS easily interpretable and incremental, we implemented an example for the RS in JavaScript (programming language highly compatible with different types of web systems). The system was register National Institute of Industrial Property of Brazil. The current version of the system can be found in a GitHub repository with a commercial license2.
Initially, following the architecture presented in Figure 2, the User Modeling Unit was implemented. Each user type is represented in an array. Another array is created to represent possible choices as to the type of recommendation (i.e., accomplishment-based or preference-based recommendation). Another array is created to represent each dominant user’s type. Finally, the last array is created to represent the recommendations possibilities (i.e., the available designs to be recommended). The User Modeling Unit is presented in the Code 1.
User Modeling Unit representation const userSchema = Schema({ userTypes: Array, choiceRecommendationType: Array, dominantTypes: Array, recommendation: Array });
To implement the recommendations (based on Santos et al. [15]), two three-dimensional matrixes were created, representing the ß-value and the P-value, accordingly to the following indexation: recommendationTable[UserType][Criterion ][Design]. In an example, considering the Code 2, to get the ß-value of the Philanthropist’s preference for the social design, we have following processing, BTable[0][1][4]. 2 Link to access the code: https://github.com/kibonusp/rsgamification-design
Content Managing Unit representation const createRecommendation = async (req, res) => {
const user = await userModel.findById(req.params.user_id ); // The User Model of a specific user is taken const accomplishment = [] const preference = [] // Two arrays are created to store the recommendations for both criteria for (userType of user.dominantTypes) {
let recommendation_based_accomplishment = maxIndexBPTable (recommendationModel.BTable[userType] [0], recommendationModel.PTable[userType][ 0]);
let recommendation_based_preference = maxIndexBPTable(recommendationModel.B Table[userType][1], recommendationModel.PTable[userType][ 1]); accomplishment.push(recommendation_ba sed_accomplishment); preference.push(recommendation_based_ preference); // Add recommendation to respective arrays } const recommendation = [accomplishment, preference]
return recommendation; }
Finally, the function maxIndexBPTable get the indexes which the value of ß-value is maximum, get the most significant p-value and provide the recommendation.
In this section, we present the work limitations, as well, an agenda for further studies. Our work contributes to the field of gamification design, providing a RS able to be adapted and plugged in general gamified systems. However, we have some limitations, which create the possibility to propose future studies that would further the knowledge in the field. Firstly, in our work we did not evaluate the RS. Therefore, future studies should evaluate the system in terms of recommendation effects (considering users preference and perceived sense of accomplishment), as well as plug the RS in different systems and evaluate its efficacy in provide the personalized gamification. Future studies also can compare the users’ experiences when using a personalized (with the RS) and a non-personalized (without the RS) version of the gamified system.
Recent studies have pointed out that using one single user characteristic to personalize gamification might not be sufficient to create a suitable gamified system for the users [30], since they have different characteristics that could influence the use of the gamified system [3]. Other aspects such as demographic information, gaming habits or personality traits can be addressed in future studies about RS, to investigate how the multiple user characteristics (besides only user traits) can be combined into different recommendations for gamification designs.
Different studies have pointed out that the use of questionnaires to assess the user type present some limitations, as for example random responses [41] or missing data [3]. Also, measuring the user type only in the first system use, might not be the best option since their profile can change over time [16], [28]. Future studies can adapt the RS to predict the user type based on user behavior data or their first answer to the questionnaire. In this way, the RS could provide recommendations that would be adapted to the user changes.
personalized and not personalized recommendation
of the context in the recommendations
recommendations
recommendations
how to create RS to automation of gamification
how to create RS to automation of gamification
In this version of the RS, we also did not consider the context of application, creating a RS that could be used regardless domain. The context can play an important role in the effectiveness of gamification [30], and prior research have indicated that studies about how the context impact the success of the implementation of gamification strategies, are a gap in the field [1], [3], [16]. Future studies can use and adapt the RS to specific domains and evaluate if the recommendations fit the user preferences, and therefore, positively affecting their behavior.
Finally, in this work we propose a RS based on gamification designs. To provide these recommendations it was possible to find only one study in the literature that related the Hexad user types with gamification designs [15]. Therefore, it was not possible to provide individual game elements recommendations for the users or create recommendations based in different studies. Since there is a large number of studies that relates the Hexad user types with the game elements individually (see [3] for a review), future studies can improve our RS using prior research to provide recommendations for individual game elements for the users. These evaluations and comparisons studies would provide the field with more evidence-based that the RS could be an option to personalize gamification. Table 2 summarize the agenda for future studies.
In this study, we propose a RS for gamification designs, capable of recommending gamification designs according to the user type. Thus, we contribute to academia and to the industry. In future work, we aim to improve the RS, provide recommendations based on other user characteristics.