A Systematic Literature Review on Enterprise Architecture Visualization Methodologies

EA (Enterprise Architecture) visualization methodologies have been explored by researchers and engineers to conduct EA modeling. The objectives of EA modeling are to clarify enterprise strategies, visualize business processes, and model information systems to manage resources, improve organization structure, adjust information strategy, and create new business value. Therefore, EA models can be broadly applied in various fields. For example, the applications include business modeling, information system architecture design, technology infrastructure configuration, software maintenance, and system security analysis. As the primary source of information, EA models are of paramount importance to researchers, architects, and developers. However, up to now, the purpose and means of these EA visualization methods have never been systematically analyzed and discussed, and a generalized EA visualization methodology with the ability to meet different demands is needed. The paper narrows this gap by conducting a systematic literature review on enterprise architecture visualization methodologies. In this study, 112 papers are retrieved by a manual search in 5 academic databases, a systematic literature review on EA visualization is explained to show a systematized category of visualization approaches, and then a general visualization approach is proposed by systematically reviewing the papers. Finally, the paper is concluded by discussing the contributions and limitations of the study.


I. INTRODUCTION
Enterprise architecture (EA) is a special field developed on the basis of the practice of information system architecture design and implementation. It is mainly used in large organizations and big projects, such as enterprise planning and government construction. Since Zachman et al. completed the pioneering work of EA in 1987, a lot of research and practice have been accumulated in this field. EA can be viewed as the means of optimizing enterprise structures, behaviors, and functions. In the development of EA, developers generally start from the perspective for overall optimum and then grasp the business strategies, the composition of the organization, business process of each department, interdepartmental collaboration, information systems, and other elements as a hierarchical structure. Through a comprehensive review of the organization, unnecessary costs can be reduced and comprehensive efficiency can be achieved with the whole organization. EA is a well-defined practice for The associate editor coordinating the review of this manuscript and approving it for publication was Laxmisha Rai . conducting enterprise analysis, planning, governance, and evaluation, with comprehensive approaches for successful development and implementation of strategies. Architects can use EA to apply architecture principles and practices to guide organizations through the business, information, process, and technology changes necessary to execute their strategies. These practices utilize the various aspects of an enterprise to identify, motivate, and achieve these changes. Although it needs a certain degree of cost in terms of time and capital, through efficiency and cost reduction, the introduction of EA often brings greater profits. Furthermore, the bigger the enterprise is, the higher the quality of EA is required.
To analyze, design, plan, and implement EA, it is usually necessary to develop EA models to realize EA visualization. Through EA modeling, the abstraction degree of the information system design can be increased, and verification at an early stage of the system development becomes possible. EA visualization enables information system designers to get a bird's-eye view for business and organization. When developing large-scale projects, EA visualization makes it easier to manage resources, strategies, risks, and business processes. Moreover, in both academic research and practical engineering, EA modeling approaches are essential to EA visualization. As a result, many methodologies for EA implementation and deployment have been explored by researchers and engineers to conduct EA visualization. The objectives of EA modeling are to clarify enterprise strategies, visualize business processes and model information systems to manage resources, improve organization structure, adjust information strategies, and create new business value. It follows that EA models have been broadly applied in various fields.
Although there are many visualization methodologies on modeling EA, however, as of yet, the application of these theories appears to be fragmented, and the approaches are rarely systematically used in empirical studies. Therefore, it is necessary to categorize the fragmented visualization methodologies. Besides, one of the reasons for the variety of EA visualization papers is the wide scope of EA. EA involves business model, requirement engineering, governance, operation, IT infrastructure, and security analysis. For each domain, there are existing visual models. It is also necessary to integrate domain-specific visual models with EA models. In this paper, the authors propose an integrated method to develop a systematic visualization approach on EA models based on a survey of existing papers on EA visualization. Fig.1 shows an overview of this paper. Concretely, the paper is structured as follows. In this section, the research background, the objectives of this research and the result are described as an introduction. Next, in Sect. II, research questions and research method for conducting the systematic literature review (SLR) [131] are explained. In Sect. III, literature selection and classification results are reported in chronological order. To identify the current situation and the future research directions of EA visualization research, the publication trends, technical characteristics, and quality of the selected papers are analyzed based on the information extracted from the selected studies in Sect. IV, V, and VI, respectively. After that, in Sect. VII, a general EA visualization method is proposed to narrow the gap between EA visualization methodologies and unsolved challenges in EA visualization research. In Sect. VIII, the related work is introduced to explain the motivation and necessity of our study in this paper. Finally, a discussion on the paper and the conclusion of the study are made in Sect. IX.

II. SURVEY METHOD
This section describes the outline of this review by defining our research questions, search database, search keywords and search process.

A. RESEARCH QUESTIONS
The research questions that we intend to answer in this study are defined as follows. The following sources have been selected to perform the survey on EA visualization methodologies. The sources are chosen as they are generally considered to be the most important and influential academic article databases in the fields of information technology and software engineering.
We use the following inclusion criteria for selecting the research studies by personally reading. Research articles that meet all these criteria will be selected for the SLR.

1) INCLUSION CRITERIA
[IC1] Studies with a clear focus on some aspects of EA visualization.
[IC2] Studies providing solutions for needs and problems of EA visualization.
[IC3] Studies subject to peer review (e.g. journal article, conference paper, workshop proceeding, and book chapters).
[IC5] Studies available as full text. We use the following exclusion criteria for eliminating undesirable research studies. Research articles that meet any of the following criteria will be filtered out.
[EC2] Studies that do not provide any conceptual, formal, or technical solution for the proposed approach to EA modeling.
[EC5] Studies in the form of white papers, tutorial papers, short papers, poster papers, or manuals. Fig.2 shows the search and screening process. The process had been completed in Oct. 2019. After identifying the articles, we analyzed them in detail. For each selected article, we recorded their bibliographical information (e.g., author affiliation, year of publication, type of publication, venue, and publisher), content-specific issues, such as the context and objective of the study, EA modeling languages, visualization tools, evaluation approach, and the number of citations. We also classified the articles according to their study operation. Last but not least, we recorded their research scope, methodology, limitations and future work.

III. RESULTS
In this section, we aim to answer the following research questions: RQ1. What are the main purposes and motivations of EA visualization? More specifically, we systematically classify the selected research papers and briefly introduce them in chronological order.
The result of the literature search is reported as follows.

A. BUSINESS MODEL
Winter et al. proposed a practicable structure for information management in hospitals, to support strategic planning and to reduce efforts for creating strategic plans [2]. Taylor and Palmer provided an overview of existing EA and then proposed an architecture for the embedded device domain [4]. Majedi and Osman proposed an architectural design model for implementing enterprise systems with the aid of SOA (Service-Oriented Architecture) [10]. The proposed model can be used to address the organizational challenges in enabling cross-communication and collaboration without changing the current ICT infrastructure. Chen et al. presented an integrated service-oriented enterprise system development framework as well as an instantiated design process model [21], to support the engineering of enterprise-wide serviceoriented systems. Antunes et al. explored the architecture context description and addressed the gap between the stakeholders' concerns and the resulting EA [24]. Clark  proposed a method using ArchiMate to model security analysis [47]. Gaaloul and Proper proposed an access control model-based approach [53] for managing organizational resources to ensure the security of EA. Sommestad et al.
presented an analysis tool called the cybersecurity modeling language (CySeMoL) to support enterprise system security managers in security analysis [54]. Korman

IV. PUBLICATION TRENDS
To grasp the current situation and unsolved challenges of EA visualization research, we analyze the selected papers and discuss the publication years, publication regions, publication venues, scholar communities, research topics and issues, and technical characteristics based on the statistical data.

A. PUBLICATION YEAR
In this section, we aim to answer the following research questions: RQ2. What does the historical development trajectory of the study of EA visualization look like? Fig.3(a) presents the distribution of publications on EA visualization methodologies over time. From the selected studies, we can observe that relatively few studies were published until 2008. Before 2008, the average number of articles published on EA visualization was 1.3. Since 2009, this number has increased to 8.9. Starting in 2009, we can see a growth in the number of studies. Coincidentally, in early 2009, the Open Group published the ArchiMate 1.0 standard as a formal technical standard. ArchiMate is a common language for describing the construction and operation of business processes, organizational structures, information flows, IT systems, and technical infrastructure. From our SLR results, ArchiMate is currently the most widely used language and framework for EA visualization, which will be further analyzed in the following section. Among the selected studies, the earliest publication on EA modeling using ArchiMate is the basic concept of ArchiMate described by Jonkers et al. in 2004. Since then the number of research articles on EA development using ArchiMate has increased substantially. From a macro point of view, another possible reason is the impressive speed at which software systems and information technology spread in the first decade of the 21st century. People's life is becoming more and more inseparable from IT. The trend of business model innovation and IT technology innovation is unstoppable. The emphasis placed on EA by government and big companies has contributed to the popularity of EA visualization research.
However, the number of research studies on EA visualization fell off a cliff in 2014 and 2016. We have also analyzed the possible causes, but unfortunately did not find any clear reasons for explaining it. Nevertheless, given the steady rate of the number of publications (6.5 papers per year) for the past 4 years, we expect that the trend would continue in the near future. Moreover, it should be noted that the search was completed in 2019, therefore inevitably, the retrieved articles of 2019 are only part of the overall studies.

B. PUBLICATION REGION AND VENUE
In this section, we aim to answer the following research questions: RQ3. In which regions do the studies of EA visualization prevail? In which venues have the selected studies on EA visualization been published?
As shown in Fig.3(b), the survey result shows that most of the authors are from the Netherlands (16.1%), followed by Germany (12.5%) and Sweden (10.7%). From the perspective of intercontinental distribution, most of the authors are from Europe, followed by North America and Asia. It should be pointed out that we have only counted the nationality of the first author of the selected papers, because if an article has ten coauthors, and they come from different countries or regions, it is meaningless to analyze all the publication regions. Fig.3(c) shows that the majority of the selected papers are published by IEEE (45%), followed by Springer (23%), ACM (14%), and Elsevier (12%). To further analyze the publication venues of the selected papers, we classified the selected research studies to assess their distribution by the type of publication (i.e. journal article, conference paper, workshop paper or chapter of book). As shown in Fig.3(d), the most common publication type is conference (60 papers, 53.6%), followed by journal articles (32 papers, 28.6%) and workshop papers (15 papers, 13.4%). On the other hand, book chapters have only 5 occurrences (5 papers, 4.5%). Overall, conferences and journals are the most targeted publication venues, testifying that the EA visualization method has become a significant research theme. Table 1 shows the publication venues that hosted selected more than 2 research studies. From the collected data, we can notice that while for journals, the selected studies mainly pertain to general software engineering or system modeling venues, for conferences and workshops there exists a prevalence of enterprise or business-related symposium. Besides, 3 selected contributions were published in a medical informatics related journal, which indicates that contributions about EA visualization are not only favored by venues related to software engineering or enterprise informatics but also interested in other practical application fields. Studies on EA modeling are spread across a large number of heterogeneous venues, with researchers paying attention to the specific EA visualization methodologies, as well as the benefits and effects of EA visualization.

C. RESEARCH COMMUNITIES
In this section, we aim to answer the following research questions: RQ4. Which institutions and researchers have made outstanding contributions in the field of EA visualization?
This section analyzes the authors of the selected papers and their relationships to investigate the ecosystem of EA VOLUME 8, 2020 visualization research. For the 112 selected studies, we identified the 415 authors of them. We used a free software called Gephi to plot the co-authorship network. It should be pointed out that we only focused on the condition of their joint research, did not discuss the relationship of citations. We first recorded and organized author information with a spreadsheet and imported it into the cartographic tool. Then we obtained the coauthored network diagram as shown in Fig.4, by applying the Fruchterman-Reingold (FR) forcedirected network layout algorithm.
The nodes in the diagram represent the researchers, and a link exists as soon as one author worked with another author on the same publication. Although Gephi supports filtering subgraphs with fewer relationships, we had visualized the entire network in order to grasp the overall situation of corelationships.
Based on the coauthored network diagram, we identified that there are several small research networks for EA visualization methodologies, but not a single big one. We also found several relatively large networks, and the scholars who act as pivots in them. The scholars who got significant achievement in the study of EA visualization are H. Jonkers from Novay information technology research institute (the Netherlands), M. Iacob from the University of Twente (the Netherlands), A. Caetano from the University of Lisbon (Portugal), and M. Ekstedt from Royal Institute of Technology (Sweden). Based on our further analysis of the co-authors of the ''pivot'' scholars, we found that research collaboration between universities and research institutions in the same region occurs frequently. Moreover, scholars who are working in both university and company at the same time, and scholars who used to work for different organizations tend to create more collaborative research relations. In addition, it is worth noting that the scholars mentioned are all from Europe. It is reasonable to consider that the active efforts of these scholars have contributed to the prevalence of EA study in the region.
In Fig.4, there are 301 nodes and 279 edges, with a network diameter of 7, an average path length of 2.733, an average weighted degree of 2.013, a graph modularity of 0.928, and a graph density of 0.006. The indicators implied that on the whole the relationship between researchers tends to be fragmented and research activities lack cooperation. For scholars committed to EA, it is necessary to pursue further mutual understanding and research collaboration in the future, as well as consistency in research methods. In this regard, we suggest that researchers should expand their knowledge accumulation and connections. Academic conferences, workshops or even forums on the Internet, may become good channels for this purpose.

D. EA VISUALIZATION THEMES AND ISSUES
To grasp the current trends and problems in EA visualization research, we analyzed the practical domains of the selected studies. Table 2 shows the statistics based on the analysis result. EA modeling is the most addressed practice (25 papers), followed by governance (17 papers) and EA evaluation (11 papers). On the other hand, service-oriented architecture and business intelligence are the least addressed practices, there are only 2 papers in which the issues are discussed respectively. Furthermore, there are also selected studies on addressing multiple issues in one publication, such as S021, which discussed Service-oriented Architecture (SOA) and business-IT alignment simultaneously. VOLUME 8, 2020  In addition to the classification of the selected papers based on research domains, it is also meaningful to detect the keywords for the selected studies. With the visualization tool TagCrowd, we created a tag cloud based on all abstracts of the 112 selected papers, except for those publications that do not have abstracts, such as book chapters. Fig.5 shows this tag cloud, which gives us an overview of the 50 most important keywords of the abstracts, and the frequency of their occurrence. It should be mentioned that we have filtered keywords like ''paper'', ''study'', ''present'', ''propose'', ''provide'', ''describe'', ''existing'' and other keywords that often appear in abstracts but have no real statistical value.
The most frequently occurred goal-related keywords are model (351 times), process (115 times), requirements (86 times), management (83 times), services (47 times), assessment (36 times), alignment (34 times), decision (32 times), and security (29 times). According to the frequency of the keywords, it can be concluded that business modeling, requirement engineering, IT management, EA evaluation, and business-IT alignment are the topics of interest. The most frequently used approach-related keywords are language (109 times), framework (73 times), case (49 times), and integration (63 times). It can be derived that the most frequently discussed technical aspects of EA visualization approaches are modeling language, EA framework, case study, and integration of EA models.
Moreover, keyword ArchiMate (62 times) also has been used frequently, and it is the only keyword of modeling languages that appears more than 30 times, indicating that ArchiMate is the dominant modeling language for EA visualization. The technical characteristics of the selected studies will be further elaborated in the next section.

V. TECHNICAL CHARACTERISTICS
In this section, we aim to answer the following research questions: RQ5: What kind of techniques are used in the study of EA visualization, and what are their strengths and weakness?
To study the technical characteristics of the selected literature, we first categorized the EA modeling operations performed by the authors in their research papers. It is necessary to explain the definition of the operations here for clarity. ''Create'' refers to EA architects creating a novel EA model or EA framework to express their propositions. ''Extract'' means EA architects acquiring the content they are interested in from the existing EA models. ''Transform'' is defined as EA architects converting an expression of EA concept into another expression according to certain principles and rules. ''Integrate'' means EA architects composing multiple EA models or EA approaches for a comprehensive one. ''Restructure'' means EA architects decomposing the existing EA model into specific ones by category, layer, or phase. ''Evaluate'' refers to EA architects assessing the quality of EA.
As we can see from Table 3, the most favored idea for the researchers was to ''create'' a new EA visualization method, with 39 papers. If the proposed method is effective and feasible, this kind of operation is the most valuable and admirable one. Next comes the ''integration'', ''evaluation'' and ''transformation'', with 19, 18, and 16 papers, respectively. The least used operation is ''restructuring'', with only 7 papers. Our analysis suggests that this might have been due  to the difficulty of the decomposition process if there are no clear restructuring requirements and principles. In addition, it is difficult for EA architects to confirm the effectiveness of ''restructuring''.
In the next step, we extracted and recorded EA visualization methods used or recommended by the researchers in these articles, including business models, EA modeling languages, and EA frameworks. Table 4 shows the EA modeling notations that have been used twice or more, in descending order of frequency.
EA models describe key elements of EA. The objective of the EA model is to capture the economic value, business process, system application, and IT infrastructure of new technology such as e-healthcare solutions. To compare the EA modeling languages, we define the key features for comparison based on the 5W1H interrogatives and the visual expressiveness levels are explained as follows.
·Who The interrogative ''who'' describes the active actors of the EA model. The active actors are the provider, partner, and customer. An example question to assess ''who'' feature is ''who are the key actors of the EA model''. The ''who'' feature of the e-healthcare solution is as follows.
Provider: E-healthcare service provider Partner: Doctor, Nurse, Medical receptionist Customer: Patient ·Why The interrogative ''why'' describes the motivation elements of the EA model. The motivation elements are the concern, goal, and value chain. The concern shows the reason to achieve the stated business goal. The goal is the ideal condition that the enterprise should reach. The value chain describes the reason why the business succeeds. An example question to assess ''why'' feature is ''Why the information system is needed''. The ''why'' feature of the e-healthcare solution is as follows.
Concern: Health, safety, efficiency Goal: Safe, convenient and worry-free healthcare service Value chain: Value exchange flow in key activities ·Where The interrogative ''where'' describes the context of the EA model. The context elements are the problem, cause, and channel. The problem can be resolved by analyzing the cause. The channel describes the means to provide customers with the product or service. An example question to assess ''where'' feature is '   1: The feature is indirectly described. 0: There is no means to represent the feature element. Among the EA visualization notations, ArchiMate, UML, SoaML, and SysML belong to architecture-based modeling notations, BMM, CySeMoL, GSN, CMMN, and URN are goalbased modeling notations, and BPMN is a process-based modeling notation. The visual expressiveness levels of Archi-Mate, UML, BMM, BPMN, CySeMoL, SoaML, SysML, GSN, CMMN, and URN are evaluated as follows.
ArchiMate provides the visual icons for business actor, goal, problem, cause, channel, asset, product, information, event, trigger, flow, business process, and customer relationship. It also provides visual nodes for the concerns, cost structure, and value chain. Thus, the quantitative visual expressiveness levels of ArchiMate are {12, 10, 12, 12, 12, 11} (see Table 6).
UML provides the visual icons for business actor, trigger, and flow. It also provides visual nodes for the asset, information, event, and process. However, UML does not provide means for representing the concern, goal, value chain, problem, cause, channel, product, customer relationship, and cost structure. Thus, the quantitative visual expressiveness levels of UML are {12, 0, 0, 6, 11, 3} (see Table 6). Similarly, the other visualization notations in Table 6 can be quantitatively evaluated. Let A and B be two visualization notations. We define A > B if the sum of the key features of A is greater than those of B. Therefore, the following equation can be derived from table 6.
The result illustrates that ArchiMate is the most powerful EA modeling notation in the visual expressiveness level comparison. Meanwhile, ArchiMate is the EA modeling language most frequently used by scholars as shown in Table 4, which arouses our attention and interest for it.
ArchiMate was originally designed in Netherland, it is now standardized by The Open Group since 2009. The latest version 3.1 of ArchiMate has been published in 2019. It provides a clear way to express the construction and behavior of business processes, organizational structures, information flows, software systems, and technology infrastructure based on the concepts of the IEEE 1471 standard. ArchiMate offers an integrated architectural approach that describes and visualizes different architecture domains and their underlying relations and dependencies. ArchiMate distinguishes itself from other modeling languages such as UML and SysML by its enterprise modeling scope. ArchiMate has the advantage of a wider scope, not only for software or system modeling but also for EA modeling. Specifically, it can be used to visualize business processes (this aspect covers the scope of BPMN), and conduct enterprise strategic planning. Table 7 shows the EA visualization frameworks that have been recommended in the selected papers. To concretely evaluate the EA frameworks, we have developed a framework for comparison as shown in Table 8. Subjective evaluation items such as understandability, complexity, and usability are omitted to ensure the objectivity of comparison, because evaluation results may conflict when the items are evaluated by different analysts. We identified thirty-six feature elements using six dimensions and six interrogatives. The feature categories are layer, model, method, governance, capability, and extensibility.
Key features of EA Framework X are defined as follows. a layer feature of X),  Table 7 can be evaluated. Table 9 shows the assessment results, and we use a radar chart (see Fig.6) to represent the data. The result shows that TOGAF is the most powerful EA framework, followed by DoDAF and FEAF.

VI. QUALITY EVALUATION
In this section, we aim to answer the following research questions: RQ6: How are the qualities of the selected studies?
To provide an indication of the quality of academic research on EA visualization, we quantitatively evaluate the quality of each selected research study in this section. We defined a set of criteria for evaluating the quality of research studies in an objective and unbiased manner. Each quality assessment question will be answered by assigning a numerical value. The grading standard is defined as: Yes = 1, To some extent = 0.5, No = 0.
It should be noted that the quality criteria focus more on the quality of the paper description (e.g., clarity, objectivity, completeness, essay structure) rather than the quality of the research itself. That is to say, groundbreaking or highlevel research does not necessarily get high scores under VOLUME 8, 2020    that most of the studies are of good quality. Table 10 presents the research studies with a quality score higher than 5.0.
On the other hand, we also analyzed the evaluation results of each quality assessment criterion. As shown in Fig.8, we found that the majority of the selected papers did well in the introduction of research background, purposes of the study, evaluation of the effectiveness of new findings, and future directions. However, in terms of EA visualization step description and discussion of the limitations of the proposed method, many studies still have room for improvement.
The quality and value of the papers can also be evaluated to a certain extent by their citations. Therefore, we extracted the citations of the selected studies. Table 11   with a citation of 0. However, we found that most (18 of 21) of the data appear in the rightmost column of Table 11. As Table 11 is arranged in chronological order, it means that most of the papers cited 0 times are published recently (2016 ∼ 2019). As we all know, the citations of academic articles generally increase over time, the papers are expected to gain more citations in the future.
Since the number of citations does not depend entirely on the quality and value of the research articles, it should be pointed out that the data shown in Table 11 only gives a statistic of citation rates, and are not meant for comparison among the selected papers.

VII. A GENERAL EA VISUALIZATION APPROACH A. OPEN ISSUES AND GAPS OF CURRENT RESEARCH ON EA VISUALIZATION
Among the selected papers, 37 (33.0%) studies have discussed their limitations and unresolved challenges, and 82 (73.2%) studies have clarified the directions for future work. Table 12 describes the limitations and unsolved challenges discussed in the selected papers. The most mentioned concern was the scope of the proposed EA visualization methods, with 10 studies. The authors of the 10 papers considered that they should try for extending the scope of their proposed EA visualization method. For example, the validity section of S111 admitted that the applicability of the proposed solution for representing variability in EA to other domains remains unclear. Furthermore, 8 papers claimed that more empirical evaluation should be conducted to verify the effectiveness of the proposed methods. 7 papers discussed that the proposed methods need to be improved in some aspects to improve the completeness of the proposed EA visualization methods. Table 12 indicates that it is necessary to develop a general method for EA visualization with a wider application scope that addresses different requirements. Because visualization methodology is the means for EA description after all, EA development is the real objectives for most EA architects. VOLUME 8, 2020 If there is a comprehensive methodology or process for EA modeling, EA architects can focus on the EA development itself without spending too much energy on EA visualization approaches. However, unfortunately we did not find such a powerful EA visualization method in the selected 112 studies. In the next part of this paper, we will propose a comprehensive EA visualization method. The inspiration for the method comes from the EA visualization operations in the selected paper in Table 4. We aim to integrate these visualization operations to propose a new EA visualization method in the next section.

B. OVERVIEW OF THE PROPOSED APPROACH
In this section, we propose a general approach to conduct EA visualization. First, we give an overview of the approach, and then we give detailed steps for application. An overview of the proposed general EA visualization approach is shown in Fig.9. The model is composed of a 6-phase process. The phases are 1) create, 2) extract, 3) translate, 4) integrate, 5) restructure, and 6) evaluate.

1) CREATE A VISUAL DIAGRAM FOR TARGET EA
Create the visual diagram for the EA object based on the enterprise scenario of the target system. The elements of objects and their relationships can be defined and visualized by using a specific common model. The diagram can also be decomposed into node elements and their interrelationships.

2) EXTRACT OBJECTS AND RELATIONS FROM THE DIAGRAM
Convert the visual diagram developed in the previous step into objects and relations by analyzing the diagram.

3) TRANSLATE THE DIAGRAM INTO OTHER DIAGRAMS
Transform the objects and relations extracted from the diagram into concepts of other EA visualization method for a new diagram based on mapping rules, so that the two diagrams have the same meaning.

4) INTEGRATE DIAGRAMS INTO ONE DIAGRAM
Integrate different EA concepts into a composite EA concept set by merging the same concepts and retaining the different concepts.

5) RESTRUCTURE THE DIAGRAM INTO COMPONENT DIAGRAMS
In the integrated model developed in the previous step, the elements of different layers are mixed together, resulting in a high degree of aggregation of the EA model. For the purpose of improving the readability and maintainability of the EA model, in this step, we restructure the integrated diagram into component diagrams for different layers (Business Layer, Application Layer and Technology Layer).

6) EVALUATE DIAGRAMS
Assess the capability of the model according to rating criteria.

C. DETAILED STEPS OF THE PROPOSED EA VISUALIZATION APPROACH
The proposed method consists of 6 phases. it should be noted that each phase can also be carried out separately to meet specific needs. EA models can be developed by performing the following steps in sequence.
[End of Phase 5] [Phase 6] Evaluate (Step6-1) Assess the capability of the developed EA models according to rating criteria such as information cover rate.

VIII. RELATED WORK
Researchers are pushing back the frontiers and opening doors to reveal why EA happens and how it works. So far, many pieces of research have studied the current situation and problems of EA research by means of literature review.
Niemi described a study that aims to explain the benefits of EA by a comprehensive literature review and a focus group interview of practitioners in 2006 [113]. As a result, a categorization of the EA benefits has been composed and analyzed.
Jørgensen summarized experiences from years of practice, research and development in the field of enterprise modeling, and points out directions for future development in 2009 [114].
In 2010, Franke et al. investigated the actual application of EA, by giving a broad overview of the usage of enterprise architecture in Swedish, German, Austrian and Swiss companies [115].
In 2014, Heyl presented a systematic review on processoriented management literature to analyze the literature in the areas of Strategic Management & Business Models (SMBM), Business Process Management (BPM), Enterprise Architecture (EA), Total Quality Management (TQM), and concluded with proposals for future research work [115]. Kakarontzas et al. identified important quality and functional requirements for smart cities by conducting a survey on the EA framework for smart cities [116].
In 2015, Rouhani et al. conducted an SLR on EAIMs, in which they assessed the current problems in EAIMs and discussed the tools used by EAIMs [117]. Rasti et al. analyzed the published EA related research papers from 2005 to 2014, to examine the status and progress of EA research, and to propose areas for future studies within the area [118].
In 2016, Santana et al. performed a literature review and create a state-of-the-art description of EA network analysis, applied measures and its main achievements [119].
In 2017, Gorkhali and Xu conducted an extensive review on 177 journal publications in the field of EA. In their paper, they have presented scholars and practitioners with a detailed overview of the available research in the field of EA [120]. Dang and Pekkola conducted an SLR to identify the major research topics and methods in studies focusing on public sector EA [121]. Their study showed that it seems that public sector EA is scattered, and there is no strong, single research stream. Moreover, there is consequently a need for more research in general, and targeted research in some specific segments. Egeten et al. conducted an SLR to analyze the components of an enterprise architecture framework to support business processes with the technology of an e-commerce system [122]. To propose a method to carry out the quality assessment for enterprise architecture models, Timm et al. suggested the EAQF (Enterprise Architecture Model Quality Framework) by analyzing related work by dint of a literature review in a design science research setting and applied it to a real-world scenario [123].
In 2018, Zhang et al. answered their research questions through the 5W1H analysis by carrying out a survey of BITA (business-IT alignment) research using EA [124]. These research questions aim to acquire a thorough understanding of BITA from the perspective of EA, to discover weak points in the status quo, and to identify future research directions. Nardello et al. developed a topic model to help structure the EA research field and enable EA to evolve coherently [125]. In this study, the authors presented about 360 identified topics in EA literature and their evolution over time. Yamamoto proposed an EA visualization method by conducting a survey on 49 publications relevant to EA [126], which is the prior research of this study. However, the number of reviewed literatures is relatively small, and the steps of the proposed method are not clearly defined.
In 2019, Gong and Janssen conducted an SLR on EA methodologies in which they revealed that EA is a broad concept that is interpreted and used in many different ways, and discussed the value of EA [127]. Zhou et al. integrated VOLUME 8, 2020 several typical EA modeling methods and proposed a visual innovation methodology based on the ArchiMate framework. Nevertheless, the scope and effectiveness of the method were not fully evaluated, and there were problems in the analysis of experimental data [128]. Bork et al. conducted a survey on modeling language specification techniques [129]. However, the study focused on modeling specification rather than modeling methodology.
In summary, although there are a lot of existing research papers reporting literature reviews on EA to date, none of them has reviewed the existing research work about EA visualization systematically and adequately. Therefore, it is necessary to make a systematic review of EA visualization methods to systematically classify the EA visualization methodologies and summarize a general EA visualization approach to meet the requirements of different purposes.

IX. DISCUSSION
We discuss the study presented in this paper from three aspects: the effectiveness, limitations, and contributions of the study and the proposed EA visualization approach.

A. EFFECTIVENESS
In this section, we verify the effectiveness of our study in this paper by answering the research questions put forward by using the 5W1H method in Sect. II. RQ1. What are the main purposes and motivations of EA visualization (why)?
We selected and analyzed 112 studies related to EA visualization, then categorized the selected research papers according to the objectives of EA visualization, and briefly described their characteristics and contributions in Sect. III.
On the whole, besides EA modeling, there are more researches on IT governance and system security analysis, and fewer researches on operation model, evaluation and verification. The reason may be that EA is more and more widely used in the information age, and the requirements for system security analysis are inevitably increasing. The study of the operation model requires researchers to grasp the whole situation of EA. For instance, in paper S48, the authors put forward a set of visual EA models to provide the overall process specification of business service, operation and management. It is not easy for EA architect to estimate or consider the cost of EA evaluation and verification. For example, the verification results acquired from the simulation environment are not convincing enough, and it is difficult for us to put an EA model into a real enterprise for verification. Moreover, the evaluation of EA presumably can only be carried out qualitatively.
RQ2. What does the historical development trajectory of the study of EA visualization look like (when)?
In Sect. IV(A), we analyzed the publishing trends of EA visualization research, including year of publication and the number of papers. We analyzed the relationship between the number of selected papers and EA modeling specification and predicted the development trend of EA visualization research in the future. However, the analysis is not in-depth enough to discuss the development of landmark papers. To improve this disadvantage is one of our future research directions.
RQ3. In which regions do the studies of EA visualization prevail? In which venues have the selected studies on EA visualization been published (where)?
The survey results in Sect. IV(B) showed that most of the authors are from the Netherlands, Germany, and Sweden. From the perspective of intercontinental distribution, most of the authors are from Europe, followed by North America and Asia.
The majority of the selected papers are published by IEEE, Springer, ACM, and Elsevier. And the most common publication type is conference, followed by journal articles, workshop papers, and book chapters. Overall, conferences and journals are the most targeted publication venues, testifying that the EA visualization method has become a significant research theme in the academic community.
RQ4. Which institutions and researchers have made outstanding contributions in the field of EA visualization (who)?
In Sect. IV(C), we analyzed the authors of the selected papers and their relations to investigate the ecosystem of EA visualization research. The scholars and institutions that we mentioned earlier have made outstanding contributions in the field of EA visualization research with their highquality publications. Besides, the researcher co-authorship network shows that cooperation in EA visualization research is prevalent in some small researcher communities, but most of the researchers come from the same institution or region. On the whole, there is a lack of connection and cooperation between scholars from different institutions.
RQ5. What kind of techniques are used in the study of EA visualization, and what are their strengths and weakness (what)?
As mentioned in Sect. IV, we found that the most frequently used EA modeling languages (top 3) are ArchiMate, UML, and BMM, which are recommended by 41, 13, and 11 papers, respectively. Moreover, to make a detailed analysis of the EA modeling notations and frameworks, we compared the frequently used EA notations and frameworks, and we claimed that ArchiMate is most suitable for EA modeling due to its completeness and powerful visual expressiveness level for EA visualization.
Additionally, considering the current development of new technologies such as AI and 3D visualization, we believe that AI and 3D visualization will also be widely used in the research and practice of EA visualization. In fact, there have been studies using related technologies, such as paper S068.
RQ6. How are the qualities of the selected studies (how)? As discussed in Sect. VI, we found that the majority of the selected papers did well in the introduction of research background, purposes of the study, evaluation of the effectiveness of new findings and the prospect of future work. However, in terms of EA visualization step description and discussion of the limitations of the proposed method, many studies still have room for improvement.
Many scholars have expressed their concerns about the scope and completeness of their EA visualization methods. Most of them have discussed the concerns as research limitations, and hope to expand and improve their EA visualization methods in future work.
Furthermore, scholars also discussed the necessity for more empirical evidence to evaluate the effectiveness of their methods. Nevertheless, it is difficult for scholars to apply their research results to practical EA projects to obtain empirical evidence. This remains unclear and becomes an unsolved challenge that should be addressed in the study of EA visualization. This may be a difficulty in the application of EA, because persuading stakeholders to use an unfamiliar EA visualization method to model EA is not easy. Although the "university-industry " relation may solve this problem to some extent, more research must be done to illustrate the necessity and benefits of EA visualization. In addition, it is difficult to objectively evaluate the cost and risks of applying EA visualization.
Although there are many visualization approaches that have been explored to carry out EA visualization, there is no systematic visualization method to integrate the fragmented existing approaches. To narrow the gap between EA visualization approaches and the unsolved challenges in EA visualization research, we proposed a general approach to conduct EA visualization and described the detailed steps for applying the approach in Sect. VII.

B. LIMITATIONS
The process of SLR can be divided into several steps: 1) identifying the research questions, 2) selecting data sources, 3) choosing search terms, 4) applying screening criteria, 5) doing the review, and 6) synthesizing the results. We review these steps to discuss the limitations of our study in this paper.

1) RESEARCH QUESTIONS
In this paper, we have chosen four questions as RQs and presented our research by answering these questions. Nevertheless, some new content can be added to enrich the completeness of the study, such as the frequently used tools or IDEs for EA visualization.

2) DATA SOURCES
We selected the research papers from the ACM digital library, IEEE Xplore, Springer Link, Science Direct, and Google Scholar. Although the databases are famous and high-level sources for computer science and software engineering, other databases, such as Scopus, can also be used to select the target studies on EA visualization.
In addition, the crawler script developed by programming languages such as Python seems to be useful for the literature search to improve the accuracy and efficiency of the search.

3) SEARCH TERMS
Although we think that we have chosen the appropriate data sources and keywords to search the target studies, there are also studies with high quality that are not in these databases, or they have not used topics, keywords, and abstracts related to EA visualization, so they cannot be selected for review. This seems to be a difficult problem that all SLRs have to face.

4) APPLYING SCREENING CRITERIA
A step of ''consulting EA experts'' can be added into the screening criteria to further screen research papers to reduce review time and improve the quality of the SLR.

5) THE REVIEW
Although we have extracted and recorded a large amount of data from the selected papers, some aspects of the discussion are not sufficient, we could have acquired more useful information from them. For example, the interaction between different studies has not been discussed, as a result, the research stream on EA visualization has not been formed.

6) THE RESULTS
The main findings for each research question can be highlighted to improve clarity. Additionally, the proposed general EA visualization approach needs a formal method for description and empirical evidence for further assessing its effectiveness. Moreover, case studies to validate the effectiveness of integrating different diagrams are needed. Although the general approach aims at solving EA modeling with different requirements, it is not applicable for all the contexts, such as the automated EA modeling mentioned in paper S038.
Resolving the challenges will become the future directions of this study.

C. CONCLUSION
To identify, classify, analyze, and evaluate existing methods for EA visualization, we reviewed the research papers on EA visualization systematically. We selected and analyzed 112 research papers, and then we categorized them according to their purposes. The contributions of our study in this paper can be summarized as follows.
• We retrieved more than 1000 research papers about EA visualization methodologies, finally selected 112 studies and conducted a systematic literature review to investigate the current status, technical characteristics, and unsolved challenges of EA visualization research.
• We compared the most commonly used EA modeling languages and frameworks in the selected papers, and discussed the gaps and open issues in EA visualization. VOLUME 8, 2020  • We found that the existing EA visualization methods can be categorized into 7 types according to their purposes, and there are 6 kinds of visualization operations. To provide a solution for EA visualization with different needs, we proposed a general EA modeling approach.
• we made a critical discussion on the paper based on the process of the systematic literature review.
It should be mentioned that our study in this paper is not a one-off outcome. We will continue to refine the SLR process and update our findings. In addition, we found some novel EA modeling concepts in the process of the SLR, such as EA modeling using VR (Virtual Reality), 3D EA modeling, etc. This will also broaden our research horizons, and provide directions for future research. APPENDIX See Table 13.
QIANG ZHI received the master's degree in information science from Hiroshima University, in 2016. He is currently pursuing the Ph.D. degree with the Graduate School of Informatics, Nagoya University, Japan. Previously, he was a Software Engineer with Japanese Software Industry. His research interests mainly include software engineering and information security.
SHUJI MORISAKI received the D.E. degree in information science from the Nara Institute of Science and Technology, Japan, in 2001. He is currently an Associate Professor with the Graduate School of Informatics, Nagoya University, Japan. Previously, he was a Software Engineer with Japanese Software Industry. His research interests include empirical software engineering and software quality.