1. Introduction
The topic and/or discipline of design is becoming very important and relatively young but has matured rapidly in the last decade with the increased use of digital phenomena in different fields [
1]. As a result, research related to design is growing exponentially. Scholars/researchers are investigating and exploring a variety of disciplines of Design research and disseminating under several of the fields (i.e., engineering, CAD, management, ergonomics, business, education, and art and design); to extend the experience and knowledge in this domain for this arena [
2].
The design was defined by a range of variety of design research definitions. Many definitions point to the importance of this subject in several design domains [
1,
3,
4,
5]. Rachel Cooper, one of the founders and editorial chairs of ‘The Design Journal’ was defined ‘Design’ in his journal publication which is called ‘Design Research Comes of Age’ as an initial trial issue published in 1997 (volume 0, issue 1) stated: “When we say ‘design’ we mean: the design disciplines covering products, places, and communication (i.e., graphic design, information design, product and industrial design, fashion and textiles, interior design and designer/maker issues), design management (design strategy, design policy, marketing and design, design and manufacturing, innovation), design theory (design methods, psychology, and design, creativity and design), Eco and environmental design, gender issues in design. We anticipate these topics will be addressed from an educational, historical, technological, or practical perspective. We believe these disciplines can provide a rich perspective, each informing and contributing to the depth and breadth of design research” [
6].
Ralph and Wand [
7] defined the concept of design after they reviewed of literature of existing definitions and stated: “The report views the design activity as a process, executed by an agent, to generate a specification of an object based on: the environment in which the object will exist, the goals ascribed to the object, the desired structural and behavioral properties of the object (requirements), a given set of component types (primitives), and constraints that limit the acceptable solutions.
The main and/or major directions in Design research are related to anyone who really is so interested in design to cultivate his/her understanding of how designers think and work [
8,
9]. Moreover, design offers designers/non-designers the opportunity to create/create effective, efficient, original, and impressive designs. In this sense, the design goes beyond a set of design tools or practical skills and is a process [
8]. Nevertheless, the design concept is expanding to include more diverse disciplines and disciplines. Therefore, it is essential to add value and quality to our lives as humans [
10,
11].
The movement of Design research as a discipline toward facing dilemmas in the coming years. Design research as a discipline and the concept of ‘Design’ has seen tremendous extension/enlargement of interest in recent years; in particular, from management, health, education, design industry, ICT and its applications, and business books as Design terms considers an interdisciplinary [
12,
13]. Moreover, the amount of Design research is growing with the increase in the number of journals, conferences, books, and magazines dedicated to design as a discipline. Design research is an interdisciplinary topic that covers various sciences (i.e., social, practical, computer, health, education, engineering, culture, history sciences, etc.). Several studies [
8,
14,
15,
16], reviewed Design research as a discipline to achieve a foundational understanding of this domain. These studies focus on a variety of topics cover ranging from investigating the main research streams on the ‘Design research’ concept. The primary research topics identified in the literature are concerned with the ability not just to build rigorous depth of knowledge but also the breadth of the discipline. Design research has consistently identified innovation, users, materials, production, etc. However, some domains in Design research (i.e., social innovation, policy design, open design, and design for specific industries and engineering, such as design for health, culture, education, IT, and design against crime are becoming very popular [
4].
The main aim of this research study is to investigate and comprehend the dominant areas or subfields of design research through an analysis of indexed keywords that were included in journal publications’ abstracts (i.e., research articles). Our analysis focused on keywords related to journal articles in design research. To the best of our knowledge, this study is the first documented effort to use a cutting-edge method termed “text mining” to examine the chosen Design research articles in a research article. This method/approach hypothesizes that its results depend on a study term’s existence (i.e., keywords), which can be more reliable than other methods/approaches that rely on contextual data or authors’ opinions. Text mining also takes into account its frequency to reflect the dominance and applicability of a phrase or keyword in the data.
This research’s motivation can easily be noticed when you think about the difference between this study and Google scholar or other search engines. Web searches, in general, may resemble text mining, but there are significant differences. Based on specific search keywords, search is the retrieval of documents or other results. The output typically consists of a hyperlink to text or information located elsewhere and a brief description of what can be found at the other end of the link. These kinds of searches are frequently carried out using search engines like Google, Yahoo, or Bing, and your company may also use an enterprise search solution.
Finding the entire existing work is the goal of using its material. The purpose of text mining is to analyze text. Instead of just looking for, linking to, and retrieving papers that contain specific data, the objective is to extract useful information. In contrast to searches, the outcomes of text mining depend on the researcher’s intended use of the data. While search functionality aids users in locating the particular document(s) they need, text mining goes far beyond search to identify specific facts and claims in the literature to create new value. The work is the first attempt to use text mining to examine research directions in design research.
This paper contributes to Design research and practice by detecting/exploring main research trends in the design research discipline and spotting light on previous and new prospective research priorities. Additionally, this research study provides perceptions/observations concerning new dimensions important in the design and associated areas. Compared to other literature methodologies, text mining with these qualities typically allows for higher applicability and validity. The Findings of this study reveal the research interest and the trends of the design research discipline. This research study utilized one of the largest samples of publications (i.e., papers). A total of 17,486 words from design journals were used in this research report. A quantitative analysis was conducted to prevent author bias when analyzing research and how we look at and analyze the data collected. Future scholars might examine and cluster the information acquired in this research study (data available in
Appendix A can be used manually). The following section presents Design research directions and the research methods conducted in state of the art. The third section illustrates this research paper’s research method and questions. Section four describes a report of results by analyzing and discussing data results. Finally, conclusions, including contributions and limitations, are stated in the fifth section. The popular design terms and abbreviations, are listed in
Appendix A and
Appendix B sequentially.
3. Research Methodology
This research paper pursued three research questions that enhanced our experience and knowledge of the design research discipline. Given below are the research questions:
RQ1: What are the major design research topics observed in the dataset?
RQ2: What changes in design studies were observed during the sample period from January 2007 to March 2019?
RQ3: What are the vital design research topics that determine the direction of future research?
Many keywords were used in the study, including articles related to design and research topics. We used a set of published keywords along with the article’s abstract, year of publication, and other information about the keywords, such as indexed keywords. The most famous studies available in the field, ranked by the index of the (ISI Collection) Web of Science website, were used in the study. Those journals’ titles were used that comprised the word ‘design’. The primary criteria for selecting the journals was the strong relation of design discipline to ICT, sciences, education, ergonomics, engineering, technology, and service/product design and development. Based on that, some journals were excluded for two reasons: first, not having been indexed by the Web of Science (ISI-core collection database) and/or considered as an Emerging Source Citation Index (ESCI), such as ‘she Ji: The Journal of design, economics, and Innovation’. However, some journals were indexed in the web of science (ISI-core collection), as they had some relation to design, such as ‘Innovation and Management Review’. However, it was excluded based on the research team’s evaluation criteria. Second, some journals were excluded that did not cover key topics, such as journals: design and culture, design for culture, design for health, and journal of design history.
On the other hand, some publications (i.e., editorial board, introduction, reviews, articles, and other sources) were published under the mentioned journals (See
Table 1). Roughly were excluded for not having the keywords that serve the research goal and/or for not aligning with the paper topic (i.e., design discipline). Therefore, these publications were omitted based on the research team’s opinion and judgment criterion. We specifically selected each journal because we wanted to assess trends in more specific areas. Additionally, by extension, this would have represented changes occurring in a broad spectrum of fields.
Figure 1 illustrates the research methodology used for the review of existing work.
We used the research published in the journals listed in
Table 1 with the corresponding study and publication years. The data covered the years 2007 through March 2019. This research report uses 3553 research articles from ten journals and 17,486 keywords. The research titles, publication year, names, and issue numbers were also added to the data collected. We also included publications from each author’s website by harvesting data mining techniques from the data set (structured and unstructured), where the keywords were designated under abstract and separated by a comma or semicolon. The research data were entered into an excel sheet (CSV) file for analysis because this is a Python language-acceptable file format. We created a Python script that extracts knowledge intelligently, automatically, consistently, and reliably from HTML and XML files to compile pertinent and important data. The data patterns that were employed to achieve the intended result were found by a smart correlation engine that was programmed. The results of the experiment demonstrated that a programmed script can mine data repeatedly at the identical levels of accuracy as a human but at highly efficient manner with privilege of ease to convert (e.g., HTML (input) to CSV (output) file format), operate, collect, and associate data.
4. Data Analysis and Discussion
In the following two parts, we will describe the two analysis directions. Using the descriptive analysis, we could relate our literature review and gain a better understanding of the domains. This allowed us to move on to the next step, which involved employing text mining techniques.
4.1. Descriptive Analysis
The research dataset inserted into Microsoft Excel was managed and used to perform a cluster and frequency analysis. In the beginning, we considered the distribution of publications per year, as shown in Figure 3. Research in design fields attracted and flourished in the ultimate few years, which appeared normal due to the vogue of the design domain and technological improvement. Furthermore, a few of the selected set of journals/publications were founded between 2007 and 2011 (See
Figure 2), which brought an essential decrease in the number of keywords’ frequencies (See
Table 2). The next step of the analysis was to produce an initial keyword distribution, where the frequency of top/popular keywords was expected (See
Table 2). Unsurprisingly, the design showed the top keyword with the highest frequency among all keywords. It seemed like the desired result based on our selection of journals.
Table 2 offers perceptions of other keywords that abundance the domain. The frequencies of the keywords are very different; the data limits the repetition of the keywords and explains the use of high-frequency associated keywords.
Table 2 represents the keywords with the highest frequency. While
Figure 3 shows the distribution of publications per year and leads to the publication distribution per journal. Keywords such as design education, design, creativity, co-design, design process, participatory design, innovation, and product design appeared as the top keywords with the highest frequency in the literature that attracted more research in the last few years. Furthermore, we found that the keywords, such as design thinking, technology education, design research, collaborative design, conceptual design, sustainability, and design cognition, drew more attention and attraction.
4.2. Text Mining Analysis (Clustering)
The document clustering text mining model was utilized to answer the study’s research questions. Clustering is a collection of data reduction techniques used to group similar observations in a dataset so that observations in the same group are as similar as possible, and observations in different groups are as different as possible.
In this study, K-means was used, which is a cluster analysis method that groups observations by minimizing Euclidean distances between them. The difference between two observations on two variables (x and y) is plugged into the Pythagorean equation to solve for the shortest distance between the two points in Euclidean distances (length of the hypotenuse).
Figure 3 illustrates clustering in more quantitative.
In this approach, clusters of keywords were created using the k-means clustering algorithm. Because it is practical, easy to use, and successful, the k-means method is frequently employed in clustering algorithms. This process does not need supervision, nor does it have predefined labels or classes. This process involves the formation of clusters determined by the similarity of keywords. A corpus summarization is provided by the clustering of algorithms that may be used to offer insight into what is contained in the corpus [
42,
43]. The “k” within the k-means clustering algorithms represents a predetermined number of clusters. The algorithm generates k random points as initial cluster centers. The algorithm then assigns each point to the nearest cluster center. A convergence criterion is then achieved by re-computing the new cluster centers until there is no more change occurring [
44]. The method illustrates in
Figure 4.
Text mining analysis can, therefore, be conducted using several tools. For instance, in this study, Python <
www.python.org> was used to perform the text mining technique. Python is an open-source language that is commonly used for text processing. Moreover, it is popularly used because its packages are highly flexible [
45]. NumPy, pandas, and sci-kit-learn python packages were used in performing k-means clustering. Besides, NLTK (the Natural Language Toolkit package) was also used to perform pre-processing tasks on the data [
46]. For instance, a ‘regular expression’ in Python was used to convert the texts to lowercase and remove punctuation and numbers [
45].
Figure 5 shows the distribution of publication per journal with number of articles in each.
In order to classify the keywords into text mining models (i.e., clusters) using the k-means clustering method, we combined all articles’ keywords into one big data file (excel sheet) and manipulated each article’s keywords one by one as one document (i.e., 3553 documents). Thus, we imported the data into Python and divided the whole dataset into three different time-span (i.e., corpora) based on their related publication year (variable; See
Figure 6). The first time span (corpus) comprised data from 2007 to 2011, and the second time-span comprised data from 2012–2015 and third time-span comprised data from 2016–2019. This division was essential to answering the research questions and analyzing the design of research disciplines for each corpus.
4.3. Clustering Results
We changed the time span and uniformed all the keywords by setting lowercase letters to produce the expected results. Any other insignificant signs that include marks, signs, words, numbers, and full stops were removed from the clusters because they had no values in the analysis. We generated equations to manipulate the compound terms, for example, to have the design process as a single term as seen; this will help research and interpretations.
TF-IDF is a weighing schema commonly and widely used during text mining research [
46]. This research involved the calculation of the TF-IDF vector for each document instead of using a simple term document frequency by keywords. TF is term frequency while IDF is the inverse of document frequency [
47]. The frequency of a particular word in a document is counted using the term frequency. The IDF’s value determines a word’s importance in a document. Additionally, inverse document frequency is determined by taking the quotient of the number of documents containing the term (DF) and the total number of documents (N) and finding its logarithm (log(N/DF)). The IDF value represents the frequency at which a word appears in a document file. This value may increase when some documents contain specific words among other documents. In this regard, the most frequent words represent each cluster.
Nevertheless, in using the k-means cluster for this study, the K-means clustering algorithm was used in defining the number of clusters (k). The best numbers of clusters for this study were specified based on a trial and error approach [
47]. This was done by comparing the values of k that were clustered against the value of k that is most applicable in each dataset. Ten clusters were used in this study. The research shown in
Figure 6 is the general distribution of publications grouped by category of time-span. About 16.9 % of the publications were made between the time-span of 2007 to 2011. Additionally, during the period between 2012 and 2015, 24.8 % of the publications were made, while 58.3 % of the publications were made in the period between 2016 and 2019. The total number of keywords for every time span is shown in
Figure 7. About 16.1% (2813) of the total keywords (17,486) related to all gathered publications fall from 2007 to 2011. On the other hand, 24.7% (4318) fell from 2012 to 2015, while 59.2% (10,355) fell from 2016 to 2019.
The frequencies of samples were decomposed for the same period, as shown in
Table 3. The design research topic is essential, especially in designing the research topic by the keywords’ distribution.
Table 3 shows the keyword distribution with general terms, where the first time span (2007–2011) included very few keywords related to the design research discipline less than ten times. This research is limited in its coverage of the first time span for the selected papers’ titles and related concepts to design topics. However, we provided a justification based on the frequency of keywords in this period compared to the frequency of total keywords, which were mentioned in the second and third time-spans.
The results of this part of the research will be discussed based on the clustering method. Following the previous study, we reduced the number of concepts in each cluster to 10 and the number of clusters in each period to 10 conducted by Abu-Shanab and Harb [
48]. We renamed each cluster for a better name based on the relationship and cluster associate information. Each cluster has been separated from the other, clearly using the unique words that best describe the cluster. The results have exposed each cluster according to the time differences, which directs future research. Some keywords closely related to the clusters were not included during the labeling of the clusters. Keywords such as technology, designs, technology, and architecture have forced us to verify the existing general keywords cautiously, which has helped the clustering process. We also came up with some restrictions to govern the blurrier of the clusters. The clusters have been estimated for a time-span of three, as shown in
Figure 8,
Figure 9 and
Figure 10. The conceptual variance represents the similarity between clusters as portrayed by the intelligent algorithm in the world of miscue [
49,
50,
51]. Our discussion has been set on the rigorous list of keywords collected from very popular designs that are highly known, as seen in
Table 2.
The previous results regarding the clustering step exposed a few research directions to carry on, where some terms appeared in the three time-spans. The following research directions existed in the three periods:
Co-creation: for instance, that work is related to design thinking, innovation, creativity, design process, and design.
Co-innovation: increasing the number in design research fields is a hot topic, and it seems very motivating as it appeared in three periods with a greater concern on collaborative design.
Ethical design: offer new insights/knowledge about the design process within design research.
Social practice design: it is essential to associate terms that focus on its adoption and other elements affecting the area—research related to participatory design, collaboration, sustainability, design innovation, and articulating design.
On the other hand, research areas/topics that began gaining popularity later, particularly during the last period (based on word frequencies), were related to design, creativity, education, research, co-design, design process, and participatory design. Identifying the relevance of examining clusters is best explained by analyzing keywords is important. The selection of keywords by the author had issues of significance, whereby it was not consistent. The limitation of keywords would have been the best if it had been concentrated on. Our results were based on the author’s arguments in a three-time period. We decided to contrite on the last two periods to bypass the limitations. (From 2012–2019), although this was likely to divest our analysis of depth over time. On the other hand, splitting the data range into last two periods (time-spans) would lead to the same limitation.
4.4. Word Frequency Distribution
A word frequency distribution was used depending on the period with the study of Abu-Shanab and Abu-Baker [
52]. Abu-Shanab and Abu-Baker [
52] estimated the frequencies of all famous words within the clusters, after which they were summated into major concepts. Their research paid attention to mobile phone purchases and use by applying mixed methods and new methodologies that aided us in this research study. In addition, the estimations that were used also focused on three periods. The data shown in
Appendix A was generated using a clustering tool. The clustering tool was used to create the data provided in
Table 3. The magnitude was assigned to the frequency regarding the total size of frequencies, after which it was compared with other terms. This would help the readers recognize the popular terms within the data. The list of a popular terms produced by the clustering process was taken and summated manually into logical terms as shown in
Table 4. Frequency distribution depends on a sample of the dataset. Text analysis was the basis of determining the frequencies and clustering words in this research.
A summation of keywords followed this step into more general dimensions (See
Figure 8,
Figure 9 and
Figure 10).
Figure 8,
Figure 9 and
Figure 10 across the period were drawn using the research directions that interest the researchers. Furthermore, we summed the new set of clusters into ten major dimensions according to each period (See
Figure 8,
Figure 9 and
Figure 10). The following trend can be seen based on this kind of analysis (i.e., text analysis). On the one hand, we experienced the thriving direction of some clusters such as design, co-design, creativity, innovation, design-thinking, participatory design, sustainability, design education, and design research.
On the other hand, we observed diminishing interest in the design strategy, perception, epistemology, philosophy of design, and pedagogy. This was unexpected considering the related literature research. Most of the terms/topics listed in
Table 4 appeared to be very motivating though some of them faded in the first period (2007–2011). However, the terms/topics in the last period appeared interesting for the researchers from 2016 to 2019. It is necessary to reveal the logic behind our classification approach, where terms (see
Appendix A) such as ‘design strategy’ might open an argument: is it a design-related issue or any design discipline? Similarly, do aesthetics only belong to usability/user experience or might they fit in a graphic design domain is also a question. Thus, these terms and others will open debates leading to enriching the topic and figures the strength of design research theories and methodologies.
5. Conclusions
The study aimed to explore the research directions with the design research topic. The study journals selected were of higher quality, congregated different keywords from different articles, and were used in the analysis. Ten journals were chosen for 3553 research articles and 17,486 keywords. This big dataset, a rich sample of keywords, was analyzed to conclude the design research’s main directions. New terms/trends were investigated as results in the design domain, attracting researchers, practitioners, and journal editorial boards. It was found that topics like co-innovation, ethical design, design thinking, co-design, creativity, social practice design, and generative methods/tools have been attracting more research. On the other hand, researchers persisted in pursuing topics such as collaborative design, human-centered design, interdisciplinary design, design education, participatory design, design practice, collaborative design, design development, collaboration, design theories, design administration, and service/product design areas. Finally, researchers and/or practitioners’ pursuit of a framework as guidelines to study the design research has faded. The design research area is guided by design theories (for researchers’ issues), design methodologies (for researchers, managerial, and/or administrative matters), and design methods/tools (for researchers’ and practitioners’ issues). A term distribution and analysis were founded based on the dataset and trend analysis (See
Appendix A). The results identified ten main clusters/categories in each period, with a few overlapping among them during different periods (See
Figure 8,
Figure 9 and
Figure 10) that govern research in other design areas. The first period (2007–2011) focused on topics like collaboration, human-centered design, collaborative design, design development, design administration, design technology and education, service/product design, creativity, generative tools, and design theories. The second period (2012–2015) focused on productive method, collaboration 1, design practice, design process, design thinking, design education, design technology, usability design, co-innovation 1, and design theories 1. The last period (2016–2019) focused on ethical design, co-innovation 1, social practice design, design thinking 1, interdisciplinary design, human-centered design 1, participatory design, co-design, creativity 1, and generative method 1.
5.1. Contributions
This research study utilized one of the most extensive samples of publications (i.e., papers). Prior research studies relied on the qualitative method and piece-by-piece assessment and had a tiny sample size (i.e., small sample size). A total of 17,486 words from design journals were used in this research report. The key categories examined and presented in this paper could determine what fields are thriving and degrading. In this study, the sample dataset underwent a quantitative analysis. A quantitative analysis was conducted to prevent author bias when analyzing research and how we look at and analyze the data collected. Future scholars might examine and cluster the information acquired in this research study (data available in
Appendix A can be used manually). Each researcher will have a unique perspective and thoughts on the subject domains. Finally, the work is the first attempt to use text mining to examine research directions in design research. Additionally, this research study provides perceptions/observations concerning new dimensions critical in the design and associated areas.
Importantly, we start classifying text as easily as possible using pre-trained BERT models. However, we found that BERT has many parameters and requires high computational resources, which are not available in our lab research. Training a model takes a lot of time and money. For future work, we will combine text mining with another existing embedding model called GloVe to accelerate the training speed of the model.
5.2. Limitations
This research paper was limited by the total number of journals used (web of science (ISI-core collection database) for data gathering. This limitation calls for a more thorough research or in-depth research projects using a more efficient approach. Design research topics are not only decisively published within the list of journals (design research is also published in chapters’ books and at proceedings’ conferences) used in this research paper (See
Table 2). Furthermore, some journals (i.e., The Design Journal and Design Issues) were excluded due to the research team’s criterion, even when they considered them significant in the design discipline. They could have led to some compelling results. In addition, the accessibility of data (i.e., keywords fetching) within each time span is considered another restriction, as the first time-span (2007–2011) comprised fewer articles than the second and third time-spans. This research paper declares this limitation important but contributes by providing an initial insight into the design research area.
Another limitation we faced in this research study was our judgment about the clusters/categories built (manual clustering as we categorized each 10 clusters into main categories based on authors’ experience, knowledge, and previous studies in design research). The types (See
Figure 8,
Figure 9 and
Figure 10) built are significant for future design research. Research studies [
53,
54] found that the design domain is derived from four dimensions based on design aspects. Those dimensions included composition, performance, experience, and communication. The measurements might be considered vague when observing the results of this research study. Therefore, we summarize that the fragmented nature of the design areas among different disciplines and the various topics that shape it (i.e., industrial, ergonomics, engineering, design and technology, ICT, and design and arts) prevent theory conceptualization.