Software Architectures for Adaptive Mobile Learning Systems: A Systematic Literature Review

Abstract

Adaptive Mobile Learning Systems (AMLSs) are technological tools that incorporate adaptive learning to generate an individual teaching–learning process for each student. Despite the proliferation of AMLS, their development is often hindered by a lack of standardization and the poor quality of existing software architectures in terms of requirements correctness and coverage. This paper presents a systematic literature review (SLR) of software architectures oriented to developing AMLS reported in the state of the art. For this SLR, we searched the ACM Digital Library, IEEE Xplore, Science Direct, Springer, and SCOPUS. Studies published in English that describe and evaluate a software architecture to develop AMLS were included. Twenty-two primary works were relevant to the present study. Based on the obtained results, we have identified key aspects that must be considered in a software architecture proposed for developing an AMLS in terms of requirements, quality attributes, stakeholders, services provided to users, views and architectural styles, components, and implementation technologies. The main finding of this work is the lack of a reference software architecture that considers all the requirements and quality attributes needed for developing AMLS. This gap hinders the effective development and standardization of quality AMLSs, suggesting a critical need for a comprehensive reference software architecture.

1. Introduction

Adaptive mobile learning systems (AMLSs) are software tools designed to provide a unique teaching experience to students and personalize their teaching–learning process [1]. In recent years, the educational paradigm shifts from traditional one-size-fits-all teaching approaches to adaptive learning have motivated the development of AMLSs [2]. These systems have allowed students to interact more with educational resources through mobile devices due to the ease of access these types of mechanisms provide [3,4].

The key benefits that AMLSs provide to students and teachers are [3,4,5] (a) individualized learning based on specific needs and preferences, (b) adaptive feedback and support in real-time, (c) flexibility and autonomy in their learning process, (d) personalized attention from teachers, and (e) the identification of enhancement areas and application of educational strategies accordingly.

AMLS have the potential to achieve academic improvement and satisfaction and provide an effective learning process for students. These aspects have motivated several works [5,6,7,8] to propose systems that implement an adaptive learning approach to (a) dynamically adjust the teaching–learning process to the characteristics of the students considering their skills or achievements obtained in the courses and (b) offer content personalization based on their preferences and characteristics identified automatically. In the context of software engineering, software architectures represent a key aspect of software development; therefore, research on the analysis, design, development, and evaluation of software architectures for AMLSs is fundamental [9,10]. This fact has inspired several works [5,6,7,8,9,10] to propose software architectures for developing AMLSs with different characteristics and functionalities. However, these works fail to fully consider the requirements for implementing adequate adaptive learning and all the requirements that arise from the system model characteristics of mobile distributed systems. This has led to several issues, such as a lack of standardization and low quality of the developed AMLS in terms of requirements accuracy and coverage. There is clearly a gap in the research that provides a comprehensive body of knowledge on the development of AMLS [9,10]. In this context, some works [5,6,7,8,9] have conducted literature reviews and systematic mapping studies of AMLSs to generate a knowledge base of the implemented adaptive approaches, machine learning techniques used, adaptation types of the proposed systems, challenges, and trends in general. However, we note that no study has presented an SLR of these works to identify the key aspects to be considered in software architectures to develop quality AMLSs, making it difficult to assess the proposed software architectures’ quality and identify trends, research gaps, and future dimensions.

Therefore, this work presents a systematic literature review (SLR) of software architectures oriented to developing AMLS proposed in the state-of-the-art. The proposed SLR aims to compare and identify critical aspects of the software architectures proposed in the specialized literature for developing AMLS. For this SLR, the search was performed using the following sources: the ACM Digital Library, IEEE Xplore, Science Direct, Springer, and SCOPUS. Studies published in English that describe and evaluate a software architecture to develop AMLS were included. Twenty-two primary works were relevant to the present study. The primary works selected in our SLR process were analyzed and compared for the requirements considered, the architectural design used, the evaluation method, and the aspects evaluated of the proposed software architecture. Derived from the results obtained in the SLR, in this work, we have identified several key aspects that must be considered in a software architecture proposed for developing AMLSs in terms of requirements, quality attributes, stakeholders, primary services provided to users, architectural views and styles, components, and implementation technologies. Our work also identified areas of opportunity, such as lacking a reference software architecture that considers all the requirements and quality attributes desirable for developing an AMLS. This gap hinders the effective development and standardization of quality AMLS, suggesting a critical need for a comprehensive reference software architecture. The paper’s structure is as follows: Section 2 describes related work. Section 3 describes the method used for the systematic review of the literature, and Section 4 reports the results obtained after the analysis of each study. Section 5 includes the discussion. Section 6 describes the threats to validity. Finally, this document closes with the conclusions and future work in Section 7.

2. Related Work

Some works [5,6,7,8,9] have conducted literature reviews and systematic mapping studies of AMLSs to generate a knowledge base of the implemented adaptive approaches, machine learning techniques used, adaptation types of the proposed systems, challenges, and trends in general. These related works are described below.

Ennouamani and Mahani [5] compared existing learning models in adaptive mobile learning (m-learning) systems. Their main objective was to identify how these models consider the characteristics of the student and their context to provide educational content adapted to each individual. The study examined various solutions in adaptive learning systems, focusing on how learner characteristics (expectations, motivation, learning styles, habits, and needs) combine with the parameters of the learner’s environment to establish an adequate learning context. Key findings of the study revealed that none of the existing systems fully integrated key learner characteristics with environmental parameters effectively to establish a complete learner context. A significant challenge identified for developers was determining the learner’s important characteristics that significantly influence the learning process and its performance.

For their part, Vallejo-Correa et al. [6] conducted a systematic review to analyze the contextual analysis approach in mobile (m-learning) and ubiquitous (u-learning) learning processes. The objective was to identify the variables used in the last decade for contextual analysis and how these are applied in said learning processes. Context modeling in mobile and ubiquitous learning analysis is based on the interpretation of three classes of variables. Internal variables include aspects such as learning styles in the educational context, while external variables cover the information recorded on mobile devices. Activities refer to connection to services and user interfaces. In recent research, considerable attention has been paid to the use of hardware variables, such as battery sensors and network connection, which are crucial for the efficient delivery of content to students. For the future, the authors propose the creation of an omnichannel architecture for educational environments in m-learning, a concept that holds immense potential for revolutionizing the way we deliver education through mobile devices. This architecture should consider academic, social, technological, and environmental variables.

Kumar and Sharma [7] presented an SLR to understand better the development of context-aware mobile learning applications (CAMLAs). These applications offer educational resources tailored to students’ specific needs and circumstances. The study identified eight types of contexts used to develop CAMLAs. These include physical condition, time, learner, cognitive, learning style, location, device, and people, with learner, location, and time contexts being the most common. Contextual information is collected using virtual and physical sensors and is mainly represented as relational or ontological data. The context detection layer captures these data, while the data layer is responsible for its storage and management. Six adaptation mechanisms were identified in CAMLAs: adaptation of learning resources, presentation of resources, feedback and support, navigation to locations, communication and interaction, and hardware resources. CAMLAs were classified into four categories: general adaptation, feedback and support, navigation to locations, and communication and interaction.

Gumbheer et al. [8] conducted a systematic review to analyze the recent progress in the research and development of Personalized and Adaptive Context-Aware Mobile Learning (PACAML). The review identified the contextual information used in PACAML studies, highlighting the importance of adapting learning according to the student’s context. Furthermore, the authors detailed the infrastructure requirements for developing mobile learning systems, highlighting the need for devices such as smartphones, tablets, and wearables to capture and process this contextual information. The application of PACAML in functional educational environments was also discussed, demonstrating its effectiveness in improving the learning experience through personalization and adaptation to the student’s context.

Finally, Kumar et al. [9] conducted a systematic mapping to provide an overview of the research on Context-Aware Mobile Learning (CAML). This approach allows the learning experience to be adapted to the educational needs and particular circumstances of the student. The main findings of the study indicate a significant trend—research in CAML has shown a marked increase, especially in the last three years. This trend reflects a growing interest and development in this field, with Asia emerging as the most active region, contributing almost one-quarter of the published studies. This observation highlights the dynamic and progressive nature of CAML in this region. Regarding publication venues, it was observed that several studies were published in high-impact journals, which underlines the quality and relevance of research in CAML. Furthermore, most of the studies focused on the development of CAML applications. However, a clear need was identified to develop more models and methods that support CAML more comprehensively, pointing to a crucial direction for future research and development in this area.

We note that no study has presented an SLR of primary works to compare and identify the critical aspects to be considered in software architectures to develop quality AMLSs, making it difficult to assess the proposed software architectures’ quality and identify trends, research gaps, and future dimensions.

3. Systematic Literature Review: Method

The systematic literature review of software architectures for AMLSs proposed in this work was carried out based on the method described by Kitchenham [11], which specifies guidelines for conducting systematic literature reviews in the software engineering area to orient researchers in the evaluation and interpretation of all available research publications concerning research questions and objectives. This method proposes three fundamental phases for the systematic review process. First, the review’s planning is carried out; this phase aims to define a search protocol for the review. The second phase focuses on the execution of the review, i.e., the search protocol defined in the previous phase is executed. Finally, the review report is performed in the third phase; all the results from the previous phases are presented. Overall, the SLR execution involves iteration, feedback, and refinement of the process mentioned above. The following subsections detail each of the phases and steps performed in our SLR based on the method proposed by Kitchenham [11].

3.1. First Phase: Planning the Review

This section describes the planning of the search protocol in the SLR proposed in this work. The search protocol defines the review’s objectives and the research questions, the search strategy, the inclusion and exclusion criteria, the criteria for assessing results, the data extraction strategy, and the synthesis strategy.

3.1.1. Review’s Objectives and Research Questions

The proposed SLR aims to compare and identify critical aspects of the software architectures proposed in the specialized literature for developing adaptive mobile learning systems (AMLSs). Therefore, to achieve this objective, we pose the following research questions (RQ):

–

RQ1: What requirements must be considered for the development of an AMLS?

–

RQ2: What software models or architectures and their main characteristics have been proposed to develop AMLSs?

–

RQ3: What scenarios have been used to evaluate the models or software architectures for AMLSs proposed in the literature?

–

RQ4: What have been the attributes or metrics that have been measured in the proposed works?

–

RQ5: What are the areas of study that the AMLS has addressed?

3.1.2. Search Strategy

The search strategy allows for the systematic completion of the review and the replicability of results. First, a search space must be delimited in the search strategy. Therefore, the search was conducted using relevant electronic databases to our research field: the ACM Digital Library, IEEE Xplore, Science Direct, Springer, and SCOPUS. In these digital databases, the terms used for this research were the following: Adaptive, Mobile, Learning, M-Learning, Context-Aware, Learning Styles, System, Software Architecture, Model, and Smart Learning. These terms were based on the keywords of the research questions proposed in this paper considering three different research domains: (1) mobile learning as the main research domain, (2) an adaptive learning system as a research subtopic, (3) and software architecture as a specified category. Therefore, the search string was as follows:

(Software architecture OR model) AND (adaptive OR context-aware) AND (mobile OR m-learning) AND (learning) AND (learning styles) AND (smart learning) AND (system).

In our research, we refer to learning styles as the diverse strategies each student prefers for learning [3,8]. The search strings were modified and refined as the research search progressed to ensure that all relevant primary works were retrieved in this study.

3.1.3. Inclusion (IC) and Exclusion (EC) Criteria

The criteria employed to select the studies for our analysis are listed below:

• Inclusion Criteria (IC)

–

Papers published in scientific journals and indexed and refereed conferences were retrieved with the search string defined in the previous section.

–

Papers published in English.

–

Papers in the domain of software architectures for AMLSs.

–

Papers that detail a software architecture and the evaluation scenario of the model or architecture.

–

Papers published between 2012 and 2023.

• Exclusion Criteria (EC)

–

Short papers, experience reports, workshop summaries, summaries, tutorials, or talks that do not provide enough details regarding the proposed software architecture.

–

Papers in languages other than English.

–

Papers that do not present a software architecture or an evaluation approach.

–

Papers published in scientific dissemination venues.

–

Papers published before 2012.

3.1.4. Evaluation of Quality, Rigor, and Relevance

The guidelines proposed by Kitchenham [11] suggest performing a quality assessment of the selected primary works. We used the quality criteria proposed by Dybå and Dingsøyr [12] to identify the scientific quality of each selected study in terms of rigor, credibility, and relevance. We evaluated these scientific quality criteria to determine the credibility and relevance of the selected primary works. To achieve this, we used the following questions proposed in [13]:

–

Is there a clear statement of the research aims?

–

Is there an adequate description of the context in which the research was carried out?

–

Is there an adequate description of the proposed contribution, method, or approach?

–

Is there a clear statement of findings?

–

Is the evidence obtained from experimental or observational studies?

–

Is the study of value for research or practice?

This quality assessment complements the IC and EC defined in Section 3.1.3.

3.1.5. Data Extraction and Synthesis Strategies

In this SLR, exhaustive research and reading processes were carried out on each of the primary works identified to extract relevant data correctly. For the synthesis process, key concepts must be organized to allow studies and findings to be compared in higher-order interpretations. In our case, the elements that have guided the data extraction and synthesis strategies are based on the characterization proposed by Plaza et al. [13]. This characterization considers aspects such as functional and non-functional requirements, quality attributes, architectural views, architectural styles, types of components, and implementation technologies.

3.2. Second Phase: Review Conduction

The second phase of the SLR comprises the following sequence of steps: (1) primary studies search and selection by applying inclusion/exclusion criteria; (2) evaluation of quality, rigor, credibility, and relevance; and (3) data extraction and synthesis. Each one is described in the following sections.

3.2.1. Primary Studies Search and Selection

The primary studies search was conducted using the proposed planning described in Section 3.1 in the ACM Digital Library, IEEE Xplore, Science Direct, Springer, and SCOPUS, resulting in 22 selected primary studies, as shown in Figure 1.

The study selection process was performed in two phases. The first phase evaluated the studies’ titles, keywords, abstracts, and conclusions, resulting in 137 non-repeated works. In the second phase, studies were selected using the inclusion and exclusion criteria outlined in Section 3.1.3. In this phase, the content of each publication was reviewed in detail to evaluate their quality and determine a rating of the influence of the research.

Table 1. Studies selected in the search and selection process.

Database	Phase 1	Phase 2	Included	Excluded
IEEE Xplore	27	14	6	21
ACM Digital Library	23	7	4	19
Scopus	25	8	4	21
SpringerLink	35	9	4	31
Science Direct	27	4	4	23
Total	137	42	22	115

also shows the results of selecting primary studies in the two phases.

In order to organize the analysis of the 22 selected primary studies and differentiate them from the rest of the references in this document, we assigned them an identifier, as seen in

Table 2. References (Ref.) and identifiers (Id) of selected studies.

Id	Ref.
P1	[14]
P2	[15]
P3	[16]
P4	[17]
P5	[18]
P6	[19]
P7	[20]
P8	[21]
P9	[22]
P10	[23]
P11	[24]
P12	[25]
P13	[26]
P14	[27]
P15	[28]
P16	[29]
P17	[30]
P18	[31]
P19	[32]
P20	[33]
P21	[34]
P22	[35]

. Concerning the years of publication of the selected studies, a variable frequency can be observed regarding the number of publications, highlighting the years 2012, 2015 2017, and 2019, as seen in Figure 2. In the most recent years, the number of studies proposing software architectures for developing adaptive mobile learning systems has decreased.

3.2.2. Evaluation of Quality, Rigor, Credibility, and Relevance

Regarding assessing the selected studies’ quality, rigor, and relevance, the questions listed in Section 3.1.4. were applied. If the paper met at least four of the six criteria, we considered the study for analysis. In this case, the 22 selected studies had sufficient quality. An example of how the assessment was carried out is presented in

Table 3. Quality questions applied to study P5.

Quality Question		CAMLES: An Adaptive Mobile Learning System to Assist Student in Language Learning
	Title
Is there a clear statement of the aims of the research?		✓
Is there an adequate description of the context in which the research was carried out?		✓
Is there an adequate description of the proposed contribution, method, or approach?		✓
Is there a clear statement of findings?		✓
Is the evidence obtained from experimental or observational studies?		✓
Is the study of value for research or practice?		✓

.

3.2.3. Data Extraction and Synthesis

For the extraction and synthesis of data from the obtained papers, the format presented in

Table 4. Data extraction format.

Obtained Data	Description
Id	Id (P1–P22)
Title	Study title
Authors	Authors’ names
Date	Publication date
Link	Paper’s DOI
Type of study	Journal or conference paper, workshop paper, book chapter
Architecture characterization	Description of the proposed architecture based on the characterization presented in Plaza et al. [13]
Summary	Broad paper’s description

was used. This form records the paper’s title, the authors, the year of publication, the download link, and a summary of the reviewed study. This form was designed to ensure we had a detailed and orderly record of the reviewed studies.

4. Third Phase: Review Results

This section presents the answers to the five research questions regarding the architectures or frameworks of Adaptive Mobile Learning Systems (AMLSs). Due to the heterogeneity of the 22 selected studies, we decided to show the analysis through a narrative synthesis.

4.1. What Requirements Must Be Considered for the Development of an AMLS? (RQ1)

Based on the analysis carried out on the 22 selected primary studies [P1–P22], the requirements recommended to be considered and satisfied in a software architecture for developing AMLSs are the following:

• Adaptation. All selected primary studies considered this requirement [P1–P22]. We identified three types of adaptation:

  ▪

  Learning path adaptation. This adaptation involves providing the student with a series of educational resources based on the learner’s characteristics, such as the goal and level of knowledge, to help the learner achieve their learning objectives.

  ▪

  Format adaptation. This type of adaptability involves the educational resource presented to the learner in an appropriate format regarding their learning style, the characteristics of the access device, or the student’s environment. For example, if a user tends to learn through videos and their mobile device has the necessary characteristics for this type of presentation format, the system will compare the educational resources that best adapt to these preferences.

  ▪

  Content adaptation. This type of adaptability consists of using information from the learner’s profile, such as their learning styles, grades, and preferences, among other things, to select educational content that can help achieve academic improvement. This can involve choosing specific topics and increasing the number of resources that will be offered, among others.

• Acquiring, managing, and using context information is another fundamental requirement in the AMLS because this type of information contributes to carrying out any class of adaptability. The context information is mainly obtained from the various sensors on the learner’s mobile device. From this information, a possible context in which a learner is immersed can be defined, e.g., location, noise, movements, light, physical activities, interactions with the application directly, or obtaining a history, among other situations. Fulfilling this requirement involves various tasks related to context information, such as sensing, preprocessing, modeling, storage, distribution, reasoning, delivery, and discovery [36].
• Obtaining, managing, and using learner information is also an essential requirement for characterizing the student and achieving any type of adaptability. This information is obtained mainly from the learner’s characteristics, such as learning styles, knowledge, behaviors, and preferences. Fulfilling this requirement involves carrying out various tasks on student information, such as processing, modeling, storage, distribution, reasoning, delivery, and discovery [36].
• Obtaining, managing, and using domain information is another crucial requirement involving the hierarchical representation of learning resources such as exams, courses, exercises, and examples. This requirement also involves establishing the difficulty levels of the learning topics addressed and the educational content formats available. Fulfilling this requirement also involves performing various tasks on domain information, such as processing, modeling, storage, distribution, reasoning, and delivery [36].
• Usability. The ISO/IEC 25000 standard [37] defines it as the ability of a product to be used by certain users to achieve specific objectives with effectiveness, efficiency, and satisfaction within a specific context.
• Generation of student history is required to achieve persistent storage of context, student, domain information, and information about the types of adaptations performed for the learner.
• Generation of the student profile is a requirement that aims to characterize and profile students based on information such as learning style, objectives, and level of studies.
• Technological resource management consists of registering, updating, deleting, sorting, and classifying information about the characteristics of the user’s access devices, such as storage size, battery levels, available presentation formats, and connections. The objective of this requirement is to use the information about the devices’ technological resources to customize the educational resources.
• Educational resources management consists of registering, updating, deleting, ordering, and classifying all educational content, including videos, courses, documents, and images.
• Heterogeneity is a requirement that arises from the diversity and differences in an AMLS regarding information sources, types of information, device hardware, and mobile operating systems, among other aspects.
• Robustness against disconnections involves providing mechanisms that ensure the operation of applications in the event of frequent disconnections [38].
• Lightness is a requirement that consists of achieving efficient use of the processing and storage resources of mobile devices during the execution of the system [38].
• Extensibility consists of generating an architecture that is open to new ways of accessing sensors and functionalities [38].
• Modularity focuses on separating responsibilities into single-purpose components [38].
• Ease of testing and maintainability. Consistency in components should facilitate the development of unit tests and maintainability [38].

4.2. What Software Models or Architectures and Their Main Characteristics Have Been Proposed to Develop AMLS? (RQ2)

This section describes the software architectures proposed in the selected primary studies in this SLR based on the following aspects: stakeholders, main services provided, architectural styles, views, or components, and technologies used to implement the proposed AMLS.

4.2.1. Stakeholders

AMLSs seek to provide students with new teaching experiences by personalizing educational content, considering their learning needs or characteristics. In this context, we can infer that AMLSs focus on users who seek to rely on a tool to improve their knowledge or academic skills in specific topics or research areas.

Therefore, we identified the different types of stakeholders that the primary studies considered in their proposals. In [P1][P2][P7][P8][P9][P11][P12][P13][P14][P16][P17] [P18][P20][P21][P22], the authors included higher-level or postgraduate students as stakeholders. On the other hand, in [P1][P3][P5][P6][P10][P17], the authors considered stakeholders who want to learn the English language from a basic level to an advanced level.

Concerning [P17], the authors considered students with learning objectives in science and Mathematics at a high school or higher level. Finally, the proposals [P9][P14][P16] focused on teaching Programming, covering topics for beginners and a medium-advanced level in object-oriented programming.

For their part, studies [P2][P4][P7][P12][P13][P15][P19][P20][P22] included teachers and administrators as stakeholders who have access to student information and manage educational resources delivered to learners. The studies in [P4][P19] also consider professionals as stakeholders.

4.2.2. Main Services Considered in the Learning Adaptation

Based on the results obtained, most of the selected primary studies [P1][P2][P3][P4][P5][P6][P9][P10][P12][P13][P14][P15][P16][P18][P19][P20][P21] presented a proposal for the design or implementation of an AMLS focused on providing content adaptation services. The studies [P2][P3][P9][P10] that performed content adaptation first requested that the users resolve a knowledge test of the subject they want to learn. Based on the grade obtained in the exam, the AMLS adapts the appropriate content to the student’s level of knowledge. Concerning the content adaptation in [P1][P5][P12][P13][P14][P15][P16], the way to provide this service is by using educational-level information of the student and their learning style. In these studies, in addition to applying a knowledge test to the learner, they provide a set of questions that will help the AMLS obtain the necessary information to create a user profile. This information adapts content when the student accesses the educational topics. Finally, in the AMLS proposed in [P4][P6][P18][P19][P20][P21], the learner is provided with a content adaptation service based on the information of the learner’s level according to their degree of studies and their learning styles, which are determined based on a questionnaire.

Studies [P3][P7][P8][P9][P10][P11][P14][P16][P17][P18][P19] present AMLSs with format adaptation services, i.e., how the presented information is adapted to the user. The approaches proposed in [P7][P14][P16] adapt the interface elements concerning the learning styles and knowledge level of the student. On the other hand, in [P3][P8][P9][P10][P11][P17][P18][P19], a format adaptation is produced in function of the types of files available in the applications, e.g., PDF, MP3, and MP4, among others. In this regard, the systems collect the learner’s learning style, access device information, and context information to identify their environment and offer the type of file that best adapts to the conditions of their context.

Finally, the proposals presented in [P5][P7][P22] provide students with learning path adaptation services based on their learning style, context, and levels of study. Cases were identified in which a questionnaire was first carried out to measure the learner’s knowledge and the learning styles used to generate a learning path. Diverse assessments are applied as the learner progresses through the course to measure their progress and adapt to the learning path.

4.2.3. Architectural Styles, Views, or Components

The selected primary studies have proposed the use of architectural styles such as components, layered, client-server, service-oriented, and N-tiers for the development of AMLS, as seen in Figure 3. The studies [P1][P2][P5][P12][P13][P16][P17][P18][P19] and [P22] used components or a layered architectural approach, which represent 45% of the primary studies selected in this SLR. On the other hand, 36% of the proposals [P3][P4][P6][P7][P8][P10][P11] and [P14] based their approaches on a Client–Server architectural style. A service-oriented architecture was used by 14% of the primary studies identified [P15][P20] and [P21]. Finally, 5% of the selected primary studies [P9] based their proposals on an N-Tiers architectural style. The architectural styles of the selected primary studies are described in detail in Appendix A.

4.2.4. Technologies Used in the AMLS Implementation

This section first describes the development technologies of the primary studies selected for this SLR.

Table 5. Technologies used in primary studies.

Study	Software	Hardware
[P1]	Web development	Cellphones
[P2]	Not implemented	Cellphones
[P3]	Mobile development (Android)	Cellphones
[P4]	Not implemented	Cellphones
[P5]	Mobile Development (Android)	Cellphones, PC, PDA
[P6]	Mobile development (Android)	Cellphones, PC, PDA
[P7]	Not implemented	Cellphones
[P8]	Web development (Mobile JQuery, Apache WebServer)	Cellphones
[P9]	Web Development	Cellphones, PC
[P10]	Mobile development (Android)	Cellphones, PC, PDA
[P11]	Not implemented	Cellphones, PDA, Tablets
[P12]	Not implemented	Cellphones
[P13]	Not implemented	Cellphones
[P14]	Mobile development (NodeJs, HTML)	Cellphones
[P15]	Not implemented	Cellphones
[P16]	Web development (PHP)	Cellphones
[P17]	Mobile development (Android, PHP)	Cellphones, PC, PDA
[P18]	Web Development (Java)	Cellphones, PC, PDA
[P19]	Web Development	Cellphones
[P20]	Not implemented	Cellphones
[P21]	Not implemented	Cellphones
[P22]	Not implemented	Cellphones

presents the software and hardware components that the primary studies used to implement their proposals.

Table 5 shows that the selected primary studies have chosen to use the following software technologies in their implementations: web development (Java, Mobile JQuery, Apache WebServer, and PHP) and mobile development (NodeJs, HTML, and Android).

4.3. What Scenarios Have Been Used to Evaluate the Models or Software Architectures for AMLS Proposed in the Literature? (RQ3)

The studies [P1][P5][P6][P9][P10][P12][P14][P15][P17][P18][P20] and [P22] evaluated their software model or architecture for the development of AMLSs based on real scenarios, case studies, or simulations. The proposed works in [P1][P5][P6][P9][P10][P12][P14][P17][P18] and [P20] developed an adaptive mobile learning system based on their software architecture to evaluate the usability of the proposed adaptation services. In these cases, the student users used the developed adaptation services through the implemented prototype in real or simulated scenarios for a specified period. Subsequently, the students answered a questionnaire evaluating the usability of the developed adaptation services. In general, the results show a positive perception and high satisfaction of the proposed adaptation services.

The proposed work in [P15] introduces a content personalization framework through cloud services. The services are personalized based on learners’ preferences and device characteristics, promoting user mobility and diversity of mobile devices, among other characteristics. This work evaluates the quality of service of a contextual information provider considering an application scenario where the required contextual information is the temperature reading at the mobile user’s location. This process results in identifying a suitable context provider for temperature provisioning considering the three attributes of quality of service: freshness, precision, and probability of correctness.

On the other hand, the proposed work in [P22] developed an adaptive mobile learning system prototype. This prototype identifies the best learning path based on implementing a shortest-path algorithm and the designed methods of a student’s learning style. Based on the t-test and one-way (ANOVA) test results, this work determined that students’ performances from the AMLS experimental group had a higher improvement rate than the control group.

4.4. What Have Been the Attributes or Metrics That Have Been Measured in the Proposed Works? (RQ4)

The selected primary studies in this SLR [P1–P22] have focused on evaluating the following attributes:

• Response time: The evaluation focused on measuring the AMLSs’ ability to support data transmission through various communication technologies.
• Usability: The capability of the prototype or application to be understood, learned, used, and engaged with by the user in specific scenarios. In the selected primary studies, the following usability subcategories were measured:

  ○

  Operability: Capacity of the developed prototype that allows the user to use it easily.

  ○

  Recommendable: This subcategory refers to how likely users are to recommend the software.

  ○

  Recognition of suitability: This subcategory measures whether the software meets users’ needs.

• Adaptation: Overall, the studies assess the content, format, and path adaptation. This evaluation and its aspects depend on their approach to achieving the adaptation.

4.5. What Are the Areas of Study That the AMLS Has Addressed? (RQ5)

The selected primary studies addressed different areas of study and educational levels, which are described below:

• Mathematics: One study [P17] has a section showing content related to mathematics, including exercises, mathematical audio, and videos, divided into subtopics corresponding to a course.
• Foreign languages: Studies [P1][P3][P5][P6][P10][P17][P20] focused on presenting courses related to learning foreign languages. These studies mainly present information on learning English.
• Computer science: Studies [P9][P14][P16] addressed educational content corresponding to programming. This content presents basic concepts of the object-oriented programming paradigm. The study in [P22] focused on a Network Security course.
• Sciences: Study [P17] focused on physics and mathematics. This proposal provided basic information on the fundamental concepts of these two branches of science.
• Others: The study in [P12] addressed adaptive learning objects in the context of eco-connectivist communities.

5. Discussion

This section presents a qualitative comparison between the 22 primary studies selected in our SLR in terms of their coverage of the desirable requirements for developing AMLSs, as seen in

Table 6. Comparison of related works.

PS	PA	CA	FA	CG	SH	SP	ERM	TRM	U	H
P1 [14]	X	✓	X	✓	✓	X	✓	X	✓	✓
P2 [15]	X	✓	X	✓	✓	✓	✓	✓	X	✓
P3 [16]	X	✓	✓	✓	X	✓	✓	✓	X	✓
P4 [17]	X	✓	X	✓	X	X	X	✓	X	✓
P5 [18]	✓	✓	X	✓	X	X	✓	X	✓	✓
P6 [19]	X	✓	X	✓	X	✓	✓	X	✓	✓
P7 [20]	✓	X	✓	✓	X	✓	✓	X	X	✓
P8 [21]	X	X	✓	✓	X	X	✓	✓	X	✓
P9 [22]	X	✓	✓	✓	✓	X	✓	X	✓	✓
P10 [23]	X	✓	✓	✓	X	✓	✓	X	✓	✓
P11 [24]	X	X	✓	✓	X	✓	✓	✓	X	✓
P12 [25]	X	✓	X	✓	X	✓	✓	X	✓	✓
P13 [26]	X	✓	X	✓	X	✓	✓	X	X	✓
P14 [27]	X	✓	✓	✓	X	✓	✓	X	✓	✓
P15 [28]	X	✓	X	✓	X	✓	✓	X	X	✓
P16 [29]	X	✓	✓	✓	X	✓	✓	X	X	✓
P17 [30]	X	X	✓	✓	X	✓	✓	✓	✓	✓
P18 [31]	X	✓	✓	✓	X	✓	✓	X	✓	✓
P19 [32]	X	✓	✓	✓	X	✓	✓	X	X	✓
P20 [33]	X	✓	X	✓	X	X	✓	X	✓	✓
P21 [34]	X	✓	X	✓	X	✓	X	X	X	✓
P22 [35]	✓	X	X	✓	X	✓	✓	X	X	✓

Where: Primary studies: PS, Path adaptation: PA, Content adaptation: CA, Format adaptation: FA, Context getting: CG, Student history: SH, Student profile: SP, Educational resources management: ERM, Technological resources management: TRM, Usability: U, Heterogeneity: H.

. Concerning compliance with the requirements, all selected primary studies consider the adaptation requirement; however, nine primary studies cover only content adaptation, three primary studies specifically carry out a format adaptation, one primary study considers only a path adaptation, seven primary studies provide a content and format adaptation, one work primarily provides students with a path and content adaptation, and only one work primarily provides students with a path and format adaptation. This allows us to conclude that no selected primary study meets the three adaptation types. These findings suggest the need to propose, on the part of the scientific community, research works aimed at developing AMLSs that fully cover the adaptation requirement by providing students with path, content, and format adaptation services.

Also, only 3 primary studies cover the requirement of generating a student history, 16 primary studies consider the requirement of defining a student profile, 20 primary studies meet the requirement of managing educational resources, 6 primary studies cover the requirement of technological resource management, 10 primary studies cover the usability requirement, and all primary studies cover the context obtaining and heterogeneity requirements. This analysis allows us to assert that primary studies only cover some requirements for developing AMLS. Therefore, these works [P1–P22] fail to fully consider the requirements for implementing adequate adaptive learning and do not consider all the requirements arising from mobile distributed systems’ system model characteristics. The lack of focus on the proposed software architectures has led to suboptimal design decisions. In these cases, inaccurate architectural design has led to a lack of standardization and low quality of the developed AMLS regarding requirements accuracy and coverage. Research groups and industry practitioners could perform the design of their software architecture based on the desirable requirements identified in this SLR to develop standardized and quality AMLSs in terms of coverage and accuracy of requirements, which will contribute to adequately implementing adaptive learning on mobile learning systems in different educational domains.

Furthermore, 45% of the selected primary works developed an adaptive mobile learning system prototype based on their software architecture to evaluate the usability of the proposed adaptation services. The proposed work in [P17] evaluated only the usability of format adaptation services. The proposed works in [P1][P6][P12][P20] focused on evaluating the usability of content adaptation services. Content and format adaptation services were evaluated in the proposed works in [P9][P10][P14][P18]. Finally, the primary work in [P5] evaluated the usability of path and content services. We note that research on usability evaluations of path adaptation services is limited; no primary work reports a usability evaluation of all adaptation types (path, content, and format adaptation). These findings highlight the need to propose research works that comprehensively evaluate the usability of path, content, and format adaptation services provided to users by AMLS prototypes developed based on software architectures. The results and comments obtained from the feedback provided by users participating in these works will contribute to improving the implementation of adaptive learning in mobile learning systems developed by research groups and industry practitioners.

One of the particularities in our search is that no primary studies were found that used architectural patterns in microservices and fog and edge computing paradigms. The architectural approach of microservices has had a considerable increase in recent years because it allows agile development, rapid deployment, improving capacity for recovery from failures, and using the appropriate technology for each microservice, considering important issues such as the required effort, implementation costs, infrastructure, and maintenance, among others [39]. Some important benefits of implementing a microservices approach are the following [40]:

• The construction of highly modular and decoupled systems that benefit from meeting quality attributes such as maintainability and scalability.
• Scalability in this type of system is benefited by a lower implementation cost, allowing the components’ independence and deployment on multiple servers.

Finally, we consider that the number of primary studies has narrowed in this field because research in recent years has focused on providing artificial intelligence techniques to improve the modules or components of path, content, and format adaptation services of primary studies [6,41,42].

6. Threats to Validity

Different factors in conducting an SLR could influence the analysis of the selected works. Tricco [43] outlines three categories of biases: identification, selection, and accuracy biases in data acquisition. In this regard, the methodological process for carrying out SLR in software engineering [11] was followed in this research. Furthermore, we implemented several measures to counter these biases. A key aspect was the collaborative and iterative nature of the research team’s work. This approach ensured robust conclusions and minimized the influence of personal biases on judgment. To mitigate selection bias, the research team planned, analyzed, and selected studies collaboratively and iteratively, making joint decisions to resolve discrepancies. These works were extracted and analyzed by at least three authors of this study to ensure robust conclusions and minimize the influence of personal biases on judgment. Finally, only studies published and indexed in reputable scientific journals and peer-reviewed conferences were considered to enhance data precision. In addition, the quality of the studies was determined, each undergoing a meticulous review.

7. Conclusions and Future Work

This research presented an SLR on software architectures aimed at developing AMLSs. The main objectives of this SLR were to compare and identify the critical aspects of the software architectures proposed in the specialized literature for AMLS development.

Based on the obtained results, we have identified various desirable requirements to be considered in the design of a reference software architecture for AMLS development: path adaptation, content adaptation, format adaptation, obtaining context, student history, student profile, administration of educational resources, administration of technological resources, usability, heterogeneity, robustness against disconnections, lightness, extensibility, modularity, and ease of testing and maintenance.

The components and layered architectural styles are mainly used to support some requirements for the analyzed studies. However, none of the selected primary works consider all the identified requirements and quality attributes proposed in the specialized literature in the design of their software architecture. Few works consider the requirement of path adaptation, and no work considers the three types of adaptation in its software architecture proposal (path, content, and format).

Only 45% of the proposed software architectures were evaluated in real or simulated scenarios that involved instantiating the architecture and involving students. In these cases, usability aspects were mainly evaluated concerning the adaptability services the developed AMLS provided. We note that research on usability evaluations of path adaptation services is limited, and no primary work reports a usability evaluation of all adaptation types (path, content, and format adaptation).

The main finding of this work is the lack of a reference software architecture that considers all the requirements and quality attributes needed for developing AMLSs. This gap hinders the effective development and standardization of quality AMLSs, suggesting a critical need for a comprehensive reference software architecture.

The research synthesis of this SLR has generated a knowledge base of the software architectures proposed in the specialized literature oriented toward AMLS development in terms of requirements, quality attributes, stakeholders, services provided to users, views and architectural styles, components, and implementation technologies. The results of this SLR can be used, on the one hand, by research groups that seek to identify relevant primary works and their advantages, limitations, and existing challenges, and on the other hand, as a guide for development teams that need to study the software architectures mainly used in real scenarios and identify the desirable requirements to be covered in the development of AMLSs.

This study has some limitations due to the nature of the research. The search words, strings, and databases may have limited the proposed SRL in this work. The search strings were limited to terms related to AMLSs, software architectures, and adaptive learning considering three research domains: (1) mobile learning as the main research domain, (2) an adaptive learning system as a research subtopic, and (3) software architecture as a specified category. In addition, the selected databases and the inclusion and exclusion criteria may, by their nature, have excluded some primary works. Finally, primary works specifically focused on providing artificial intelligence techniques to improve only the components or mechanisms that support AMLS adaptive services or learner models in AMLSs were not reviewed.

In future work, we propose to conduct a detailed analysis of the usability metrics, quality attributes, and evaluation methods applicable to content, format, and path adaptation services in AMLSs and map the use of machine learning in AMLS, elucidating the benefits and open challenges. In addition, we propose the development of a reference software architecture that considers all the desirable requirements for developing quality AMLSs.