Decoding educational augmented reality research trends: a topic modeling analysis

Abstract

This study aims to examine the temporal evolution and changes of research interests and trends in the educational augmented reality (AR) literature. To this end, 3718 articles published in the 2003–2022 period and indexed in the Scopus database were analyzed through machine learning-based semantic topic modeling and descriptive analysis. The findings indicate a notable upswing in studies on educational AR, particularly since 2015. The articles were categorized into eleven primary themes through topic modeling analysis. The three most prevalent topics in terms of volume are “Augmented Reality in Education and Cultural Heritage”, “Medical Education and Patient Care”, and “Enhancing Safety and Information in Food Consumption”. Observations across different times reveal that “Augmented Reality in Electrical and Electronic Systems” and “Gesture-Based Instruction and Maintenance” were studied in the initial periods. Since 2015, there has been a notable increase in applications falling under the “Serious Games” category. The least voluminous and slowest-evolving topics are identified as “Serious Games for Children with Autism Spectrum Disorder”, “Augmented Reality in Chemistry and Biology Laboratories”, and “Augmented Reality for Safe and Efficient Driving”. Considering the recent momentum gained by these topics, it is anticipated that they will become popular topics for future studies. This study represents a significant milestone as the first and most comprehensive research using machine learning in its field, not only explaining the current state of the field but also providing valuable information for future research efforts.

1 Introduction

Augmented reality (AR) is defined as the enrichment of the live camera image obtained from the real environment by adding various elements such as pictures, texts, and animations to serve a specific purpose (Doerner et al., 2022). AR superimposes virtual objects onto the real world, allowing users to view both simultaneously (Azuma, 1997; Doerner, et al., 2022). AR enriches reality rather than replacing it completely (Azuma, 1997; Carmigniani et al., 2011; Wither et al., 2011). AR systems exhibit three primary attributes: the fusion of real and virtual elements, real-time interaction, and three-dimensional recording (Azuma, 1997). The technological requirements of AR are demonstrators, tracking, registration, and calibration (Azuma et al., 2001). The technologies that fulfill these functions have changed and evolved over time (Azuma, 1997; Carmigniani et al., 2011; Siriwardhana et al., 2021). Today, users can view and interact with AR applications on mobile phones or tablets. With the technological developments of mobile devices, AR applications have become popular applications that reach more audiences (Shang, Zakaria, & Ahmad, 2016). In addition, the popularity of AR applications has increased with the introduction of glasses such as the Microsoft Hololens. AR seamlessly integrates with information and communication technology (ICT) tools to enhance user experiences in diverse applications (Radosavljevic et al., 2020). By leveraging the power of cloud computing, AR applications can access and display real-time data, 3D models, and interactive content (Kerawalla et al., 2006). The synergy with the internet of things (IoT) facilitates the integration of AR with connected devices, enabling users to receive contextual information about their surroundings, while artificial intelligence enhances AR capabilities through object recognition and natural language processing (Sahu et al., 2024; Syahidi et al., 2021). This collaborative use of AR and ICT tools leads to innovative solutions, spanning education, healthcare, manufacturing, and beyond (Raja & Lakshmi Priya, 2022).

1.1 Related work

The use of AR is increasing in many fields, such as education, health, engineering, marketing, and entertainment. In the field of education, there are many applications at the high school and university levels (Jamali et al., 2015; Moro et al., 2021; Tarmizi & Yasak, 2022). Numerous research studies indicate that educational AR applications positively impact the learning process (Akçayır & Akçayır, 2017; Elmqaddem, 2019; Garzón & Acevedo, 2019). One of the leading types of educational AR applications is AR software created by enriching printed books with digital content (Lytridis & Tsinakos, 2018). According to studies conducted with AR book software, these applications are useful in terms of understanding concepts, improving learning, and motivating students (Cheng & Tsai, 2016; Dünser et al., 2012). A study based on the comparison of AR with traditional instruction in teaching English vocabulary also supports the positive results. According to the results of the study, AR applications significantly increased first grade students’ motivation and performance in learning English vocabulary (Lai & Chang, 2021). The results of another AR intervention on magnetic fields with eighth grade students were also very positive (Cai et al., 2017). In this particular application, the control group employed conventional methods, while the experimental group utilized motion detection software based on AR to grasp the concepts of magnetic induction lines. The outcomes indicated that the AR-based application facilitated an intuitive comprehension of magnetic field concepts among the experimental group students, enhancing their learning objectives and aiding in better retention of the material. In a study with pedagogy students, Keller’s Instructional Material Motivation Scale (IMMS) was used to determine the degree of motivation to use AR-enriched lecture notes in the classroom (Cabero-Almenara et al., 2019). Motivation and its impact on academic performance were assessed through an AR application designed for Education Technology (ET) and Information and Communication Technologies (ICT) courses. The result of the study showed a strong correlation between students’ motivation to use AR-enriched lecture notes and increased academic performance. According to the results of another study investigating the 10-year impact of AR applications in education, AR positively improves students’ performance (Chang et al., 2022). In this context, performance is defined by the application of knowledge and skills acquired by students to real-world scenarios.

While many studies have found positive results regarding the use of AR in education, there are also some studies reporting no positive effects (Henssen et al., 2020). According to Sommerauer and Müller (2018), when it comes to long-term learning, AR software does not have a positive effect compared to traditional materials. Some studies even mention various challenges of working with AR (Garzón et al., 2019). One of these challenges is the complexity of using AR, especially for younger age groups. The use of AR can be a problem for users with limited technology skills due to reasons such as the concept of multisensory (Tzima et al., 2019). It is also stated that it causes student distraction and increases cognitive load. Billinghurst and Duenser (2012) mention that if teachers do not have sufficient knowledge about the use of AR in the classroom environment, the problem of not using these applications with full efficiency arises.

AR continues to be widely used in education, prompting numerous studies to evaluate its impact. Various methods, such as systematic literature reviews, meta-analyses, and bibliometric analyses, are employed to comprehensively examine the field. A systematic review of AR studies in medical education from 2000 to 2018 highlighted that many AR applications had low quality due to the absence of validity assessments, subpar study designs, limited sample sizes, and inconclusive results (Tang, Cheng, & Greenberg, 2020). A comprehensive review of 61 publications from 2012 to 2018 found that AR has a moderate impact on learning effectiveness, with key advantages being learning gains and motivation (Garzón et al., 2019). Another study examined 50 studies from 2008 to 2018, focusing on AR’s role in meeting the needs of all students, including those with disabilities. The results showed that AR attracted students with special educational needs, offering benefits such as increased motivation, interaction facilitation, and support for short-term memory (Quintero et al., 2019).

A meta-analysis of 46 publications assessing AR’s use in education in terms of pedagogical approaches revealed a moderate influence on students’ educational achievements. Cooperative learning was identified as the prevailing instructional strategy in AR settings (Garzón et al., 2020). A bibliometric analysis spanning a 25-year period (1995–2020) examined 3475 articles, highlighting the yearly expansion of scientific research, key contributors, prevalent topics, and outcomes of co-citation analysis. The USA, Spain, and Taiwan were identified as leading countries in scientific production, with Hwang being the most productive and cited author (Avila-Garzon, Bacca-Acosta, Duarte, & Betancourt, 2021). A content and bibliometric analysis of 62 studies on AR in science education from 2013 to 2018 noted Azuma, Dunleavy, and Klopfer as the most cited authors. The most analyzed variables were “Learning/Academic Achievement”, “Motivation”, and “Attitude”, with mobile applications and marker-based materials on paper being the most preferred material types for AR (Arıcı, Yıldırım, Çalıklar, & Yılmaz, 2019).

1.2 Gap in literature, motivation, and purpose of this paper

Each of the studies summarized in the previous section is an important study that shapes the field. Despite these contributions, such studies often fail to reveal underlying semantic patterns, research interests, and field trends (Gurcan et al., 2021; Ozyurt & Ozyurt, 2023). Due to the nature of these analyses, the number of textual sources that can be addressed is limited, or interpretations may be biased (Avila-Garzon et al., 2021). Bibliometric analysis is a rigorous method often used to analyze large volumes of data. While addressing the periodic differences of a field, it sheds light on important parameters emerging in that field (Donthu et al., 2021). However, this method focuses on certain criteria such as publication volume, citation status of authors, co-publication network, and trending topics (Arıcı et al., 2019). As a matter of fact, studies on educational AR literature have revealed the positive and negative aspects, advantages, challenges, effectiveness, and efficiency of this field, while also revealing its descriptive characteristics. However, these studies are insufficient for revealing the hidden patterns of the literature and conducting semantic analyses. Hence, it is crucial to undertake an in-depth textual examination of the utilization of AR in education to uncover upcoming research directions by addressing potentially overlooked study areas. Furthermore, the existing literature on AR in education provides valuable insights into its effectiveness and applications, emphasizing the need for a comprehensive understanding of the field. Yet, a gap in knowledge exists regarding the systematic analysis of topics and trends in AR research.

The existing literature on AR in education has provided valuable insights into its effectiveness and applications but lacks a systematic analysis of topics and trends. While studies have addressed the positive and negative aspects, advantages, challenges, effectiveness, and efficiency of AR, they are insufficient for revealing hidden patterns and conducting semantic analyses. Therefore, there is a gap in knowledge regarding a comprehensive understanding of AR’s impact on learning outcomes and its varied applications in different educational contexts. To address this gap, an in-depth textual examination using topic modeling is crucial to uncover emerging themes and inform future research directions in AR education. A topic modeling study in this domain would serve as a crucial endeavor, enabling the identification of emerging themes, key contributors, and evolving areas of interest. Such an analysis is essential to guide future research directions, inform educational practices, and address potential gaps in the current understanding of AR’s impact on learning outcomes and its varied applications in different educational contexts. In recent times, the method of topic modeling has gained prominence in research endeavors seeking a comprehensive understanding of various fields. This method aims to extract concealed semantic features within the literature, thereby unveiling research interests and trends. Indeed, numerous studies of this nature can be found in the existing body of literature. Some of these are as follows: smart city (Park & Lee, 2019), software engineering research (Silva et al., 2021), using artificial intelligence in education (Chen et al., 2022), blended learning (Yin & Yuan, 2022), healthcare operations and supply chain management (Ali & Kannan, 2022), educational data mining (Ozyurt et al., 2023), gamification (Ayaz et al., 2023), using technology in mathematics education (Hwang et al., 2023), and enriching education with virtual reality (Ozyurt & Ozyurt, 2023). In these studies, the research interests and trends of the relevant field were revealed through topic modeling analysis. It is, of course, possible to increase these examples. Topic modeling analysis has become a popular method for identifying research interests and trends in recent years. This approach allows for the discovery of latent patterns within large volumes of textual resources. In this study, topic modeling analysis was utilized to identify research interests and trends in educational AR literature (Negara et al., 2019). Machine learning-based topic modeling offers significant advantages over manual analysis in inferring semantic analysis from large datasets. These methods can effectively detect hidden relationships and meaningful patterns by automatically learning complex structures in large datasets. This allows for obtaining a more comprehensive and general understanding from large datasets, offering a scalable and rapid analysis process while saving the time and effort of human analysts (Asmussen & Møller, 2019; Ozyurt & Ozyurt, 2023). This technique relies on text mining and natural language processing (Debnath & Bardhan, 2020).

This study reveals descriptive features in the AR literature and unveils research interests, possibilities, and future visions. Since it is the first study to apply topic modeling analysis to educational AR literature, it is expected to make a significant contribution to the existing literature. In addition to determining the research interests and trends of the educational AR literature, the study will also make inferences about the direction in which the field will evolve in the near future. This will provide valuable insights for both researchers in the field and policymakers. The following research questions were sought to be answered:

• RQ₁: What is the distribution of articles in the field of educational AR by year?

• RQ₂: Which authors, countries, and journals are prominent in educational AR articles?

• RQ₃: What are the emerging topics in educational AR articles, and what is the trend of these topics?

2 Methodology

This section presents the framework of the research. In this study, descriptive analysis and topic modeling were combined to extract objective sensory feature characteristics (Kemp, Hort, & Hollowood, 2018). The analysis utilized Latent Dirichlet Allocation (LDA) in conducting topic modeling to unveil concealed topics. Topic modeling is an analysis method that uses automated tools to reveal, understand, and explain the thematic structure of large amounts of data (Kherwa & Bansal, 2019). LDA, a topic modeling method, is a generative probabilistic algorithm based on unsupervised machine learning (Asmussen & Møller, 2019; Blei et al., 2003; Jelodar et al., 2019). LDA-based topic modeling generates themes based on word co-occurrence (Jacobi et al., 2016). In this context, LDA-based topic modeling analysis is frequently preferred and used as a semantic content analysis tool to identify research interests and trends in any field (Ekin et al., 2023; De Mauro et al., 2018; Gurcan, Ozyurt, & Cagiltay, 2021; Ozyurt & Ayaz, 2022; Özköse et al., 2022). The framework of the study is visualized in Fig. 1.

Fig. 1

Stages of the methodological framework of the study

As seen in Fig. 1, the study consists of three main stages: The first stage is “data collection, preprocessing, and corpus creation”. The second phase is the “implementation of LDA-based topic modeling and descriptive analysis”. The last stage involves “exploration and naming of topics, visualization and presentation of findings”. These stages are detailed under subheadings below.

One of the most important stages in literature reviews is the process of creating an empirical corpus. In such studies, the Scopus database (which covers many databases including Web of Science (Elsevier), IEEE, Springer, etc.) produces the most results and is frequently used (Gurcan et al., 2021; Mongeon & Paul-Hus, 2016). Scopus is a multidisciplinary database that covers a wide range of scientific literature. This database hosts over 7,000 publishers, 27,000 journals, and over 80 million articles covering many disciplines, including science, technology, medicine, social sciences, and the arts. It is considered the largest database in the world and offers a wide perspective to researchers with its constantly updated scope (URL-1 n.d.). For this reason, the Scopus database was deemed necessary and sufficient to access the majority of articles in the field. The query criteria were determined by utilizing the research in the literature (Arıcı et al., 2019; Avila-Garzon et al., 2021) and the opinions of two field experts. In this context, the inclusion and exclusion criteria in Table 1 were created:

Table 1 Inclusion and exclusion criteria in the context of the research problem

Considering the criteria in Table 1, the last twenty years of educational AR literature were included in the scope of the research (the first query was run, and it was seen that there were a total of 25 articles before 2003; thus it was decided that the last twenty years would be sufficient for the integrity of the study). Accordingly, the following final query was created to obtain all English-language journal articles published between 2003 and 2022 (it was deemed appropriate to include “deep learning” and “machine learning” articles in the exclusion criteria since they were not found to have educational content):

TITLE-ABS-KEY ( “augment* reality” AND ( “education” OR “teaching” OR “learning” OR “instruction” OR “training” ) AND NOT ( “deep learning” OR “machine learning”) ) AND PUBYEAR > 2002 AND ( LIMIT-TO ( SRCTYPE , “j” ) ) AND ( LIMIT-TO ( PUBSTAGE , “final” ) ) AND ( LIMIT-TO ( DOCTYPE , “ar” ) OR LIMIT-TO ( DOCTYPE , “re” ) ) AND ( EXCLUDE ( PUBYEAR , 2023 ) ) AND ( LIMIT-TO ( LANGUAGE , “English” ) )

The search was executed on April 20, 2023, resulting in the retrieval of a total of 3718 articles, comprising 3305 research articles and 414 review articles. Information such as the publication year, subject area, and journal details, as well as the title, abstract, and keywords of these articles, was subsequently incorporated into the dataset.

The initial two research questions, RQ₁ and RQ₂, were addressed by analyzing the descriptive characteristics of the data collected. This involved visualizing the numerical data using graphs and tables and uncovering the descriptive features. To answer RQ3, which is the focus of the study and makes the study noteworthy, LDA-based semantic topic modeling analysis was carried out. In order to perform LDA analysis, the data must be preprocessed and cleaned (Asmussen & Møller, 2019). Thus, the raw data is transformed into a clean dataset that can be analyzed. In the data cleaning phase, we try to remove as many textual language-related factors as possible. This facilitates the computer processing of natural language texts (Ni Ki et al., 2022; Tong & Zhang, 2016). The stages of data cleaning are as follows:

• Tokenization: When you input texts, they will be automatically converted to lowercase, and any special characters and punctuation marks will be removed.

• Lemmatization: Words are normalized and stripped of their affixes.

• Extraction of stop words list: Removing words and stop words (a, an, is, the, of, for, etc.) that do not make sense in the text.

Upon the conclusion of the data cleansing procedure, the title, abstract, and keywords were amalgamated to construct a corpus suitable for topic modeling analysis. A preliminary examination exposed the ubiquity of the word “augmented” and other query-related terms across nearly all topics. Considering the corpus’s direct association with this field, “augmented” and “augmenting” were incorporated into the stop words list to mitigate the recurring inclusion of these terms in the topics.

2.1 Implementation of LDA-based topic modeling and descriptive analysis

The Python library Gensim was used to perform topic modeling analysis with the LDA-Multicore algorithm (URL-2 n.d., Wicke & Bolognesi, 2020). In order to fit the LDA-based topic model to the pilot analysis, we first determined the optimal values for its parameters. The initial values of the parameters α, which determines the topic distribution in the documents, and β (also known as eta), which determines the word distribution in the topics, were set as [‘symmetric’ ‘asymmetric’] for α and [‘symmetric’] for β. In addition, the hyperparameters of the model (taking into account the 4-core computer on which the code was run) were set to “workers = 3, random_state = 42, passes = 20”. For the fitting of the model, an iterative and heuristic process is proposed in the literature (Goedecke, 2017; Wienler, 2023). To determine the ideal number of topics (K) in LDA analysis, the model was created for all K values between 5 and 35. “c_v coherence measure” was taken into account in determining the appropriate number of topics, and the c_v value was calculated for each K value. The c_v coherence value takes a value between 0 and 1, and values close to 1 indicate high coherence (Lassche, 2022). As a result of the analysis, it was seen that the model with the K = 11 topic had the highest c_v coherence value (c_v = 0.4361) for the values of α and β binary [‘asymmetric’, ‘symmetric’].

2.2 Exploration and naming of topics, visualization and presentation of findings

The pyLDAvis tool was used to name the topics. PyLDAvis, a web-based visualization tool, was used to visualize and name the topics (Onah & Pang, 2021). The λ value indicating the priority order of words within topics was taken as 0.6 (Sik et al., 2023; Wu et al., 2020). Figure 2 shows a screenshot of pyLDAvis.

Fig. 2

PyLDAvis screenshot where the topics are visualized

As seen in Fig. 2, three different clustering among the topics are noteworthy. The distance between topics is also an indication that those topics have a few common words. Accordingly, while Topics 1, 2, and 6 formed a cluster among themselves, Topics 4 and 7 formed a cluster among themselves, and Topics 8 and 10 formed a cluster among themselves. These clusters show the abundance of common words among the topics. While naming the topics, help was received from two field experts. Topics were named together with field experts, taking into account the terms that make up the topics and the order in which the terms form the topic. In addition, the researchers, along with two domain experts, assessed the significance and coherence of the eleven topics derived from the topic modeling process. The topics were labeled according to the terms encapsulated within them, and the top 15 terms, exhibiting the highest frequency, were enumerated for each topic.

Once topics were named, the percentage of topics per document, the internal distribution of terms within each topic, and the overall distribution of topics across all articles were calculated. Using these distributions, changes in topics over time were calculated, leading to the identification of trends in topic development.

3 Findings

The findings obtained from the study are organized under two subheadings: descriptive analysis findings and topic modeling analysis findings. Under the first heading, findings on the number of articles by year, prominent authors, journals, and countries in this field are presented in tables and graphs. Under the second heading, the findings from topic modeling analysis, including changes over time and comparisons to other topics, are presented in tables and figures.

3.1 Findings on descriptive characteristics

In the study, findings regarding the descriptive characteristics of the educational AR literature in the last twenty years (including the 2003–2022 period) were obtained. These findings are taken directly from data produced by the Scopus database. That is, when a query is made in the Scopus database, a number of descriptive characteristics such as the most productive author name, subject area, country/territory, journals, affiliation and document/source types are listed in the left menu as a result of the query. The most productive authors, countries and journals were taken from this information and presented visually in the following sections.

Throughout this time frame, a total of 3718 articles (comprising 3305 research articles and 413 review articles) were identified. To illustrate the variation in publication numbers over the years, both the annual publication figures and the corresponding slope graph are presented in Fig. 3.

Fig. 3

Number of publications and slope graph by year

As can be seen in Fig. 3, the annual number of publications, which is shown in single digits in the early years, shows a smooth increase in general. In particular, while this increase was slow until 2015, there was a rapid increase in the number of publications after 2015. Figures 4 and 5 show the top ten authors and countries with the most publications in this field, respectively.

Fig. 4

Top ten most prolific authors in the field of AR and number of publications

Fig. 5

Top ten countries and number of publications according to the origin of publications in the field of AR

As seen in Fig. 4, the most prolific authors in this field are Billinghurst, M. (from Australia), Mantri, A. (from India), and Ong, S.K. (from Singapore). Figure 4 shows that the United States (n = 731) is the origin of most publications, followed by China (n = 306) and Spain (n = 278). It has been noted that the leading ten nations boasting the greatest publication counts hail from varied geographical areas. The specifics regarding the ten journals exhibiting the highest publication numbers in this particular domain can be found in Table 2.

Table 2 Top ten journals with the highest number of publications in the field of AR and the number of publications

Table 2 illustrates that “Applied Sciences”, an open-access journal, has the highest number (n = 78) of publications in this field. It is worth noting that among the top ten journals with the highest number of publications, there is a mix of open access and scientific journals related to the field, along with a significant number of educational sciences/technologies journals.

3.2 Findings about topic modeling

The analysis utilizing LDA-based topic modeling revealed eleven primary topics within the AR corpus, as outlined in Table 3.

Table 3 Names of topics obtained from topic modeling analysis

Table 3 displays a roster of names derived from the initial fifteen terms constituting the topics. Appendix-A encompasses the topics, along with their respective first fifteen terms and the percentage representation of each topic, organized by volume ratio. Moreover, Appendix-B provides insights into the distribution of publication numbers per subject across different years, the overall publication count, and the acceleration values (calculated using the Excel acceleration formula).

3.2.1 Findings about volume and acceleration of the topics

Initially, Fig. 6 presents the topics arranged by their volume ratios, incorporating acceleration values. As depicted in Fig. 6, the sequence of volume ratios and acceleration values for the topics is nearly identical. Only the acceleration values of Topic_8 (Acc = 0.45) and Topic_9 (Acc = 0.46) break this order, but the difference between the values is very small and does not change the general situation.

Fig. 6

Volume ratios and acceleration values of the topics

As seen in Fig. 6, the top three most voluminous topics (Topic_1, Topic_2, and Topic_3) are “Augmented Reality in Education and Cultural Heritage”, “Medical Education and Patient Care”, and “Enhancing Safety and Information in Food Consumption”, respectively. Considering the volume and acceleration values of the topics, three classifications can be mentioned, as shown in Fig. 6. These are high volume/very fast topics, medium volume/fast topics, and low volume/slow topics. The three topics with the highest volume are also the ones experiencing the most rapid acceleration. In terms of volume, the least voluminous and slowest topics are “Serious Games for Children with Autism Spectrum Disorder”, “Augmented Reality in Chemistry and Biology Laboratories”, and “Augmented Reality for Safe and Efficient Driving” (Topic_9, Topic_10, and Topic_11), respectively.

3.2.2 Findings about the changes in the topics over time

For a comprehensive examination of topic evolution spanning a 20-year timeframe, we segmented it into five intervals, each spanning four years. This approach facilitated the observation of variations in both the publication count and volume ratios of individual topics over time. The breakdown of the number of publications for each topic in each period is presented in Table 4.

Table 4 Number of publications on the topics in periods

Through the data in Table 4, the volume ratio of each topic was determined horizontally, and the acceleration of these ratios was calculated. For example, it can be calculated that the volume ratios of Topic_1 in the periods are 0.30%, 1.87%, 8.56%, 22.05%, and 67.22%, respectively (number of publications on the topic in the period/total number of publications on the topic). The sum of the percentages in each row can be verified as 100%. Computed using this methodology, Table 5 provides the volume ratios and acceleration values for all topics, arranged in order of acceleration.

Table 5 Volume percentages of the topics in periods ordered by acceleration

As can be seen in Table 5, the top three topics with the highest number of publications over time are “Augmented Reality in Chemistry and Biology Laboratories”, “Medical Education and Patient Care”, and “Serious Games for Children with Autism Spectrum Disorder”. Among these topics, “Medical Education and Patient Care” (Topic_2, Acc = 16.91) is also a “high volume/very fast topic” while the other two are in the “low volume/slow topic” category.

Another deduction drawn from the information presented in Table 5 is discerning the periods when the topics receive more extensive study. By making a row-based reading on Table 5, a significant increase in the volume percentage of each topic was observed in the transition between periods. The point at which there are sharp increases in the transitions between periods is accepted as an indicator that the relevant topic has started to dominate. For example, it can be said that the topic in the first row (Topic_10) “Augmented Reality in Chemistry and Biology Laboratories” has been predominantly studied since 2015. In this way, all topics were tracked, and the timeline in Fig. 7 was created.

Fig. 7

Timeline showing when each topic starts to emerge predominantly

As can be seen in Fig. 7, “Augmented Reality in Electrical and Electronic Systems” and “Gesture-Based Instruction and Maintenance” were studied relatively early on, while “Enhancing Safety and Information in Food Consumption” was studied mainly in 2007. Similarly, “Medical Education and Patient Care” started to be studied predominantly in 2011, while “Serious Games for Children with Autism Spectrum Disorder” began to gain prominence in studies in 2015.

3.2.3 Findings on the change of topics over time compared to other topics

To gauge the intensity of studying different topics, not only were the topics studied over time, but their intensity in comparison to other topics was also evaluated. To achieve this, data on the number of publications related to the topics mentioned in the previous section during different periods (as shown in Table 2) were utilized. By reading Table 2 column-wise, the ratio of the volume of the relevant topic in each period was calculated (number of publications in period i for each topic/total number of publications in period i). In this way, for each of the periods, the intensity of study of all topics compared to other topics in the relevant period was determined. Table 6 was obtained with this calculation. The sum of the percentages in each column in Table 6 can be verified as 100%.

Table 6 Order of the topics’ volumes in each period compared to the other topics according to acceleration

Table 6 reveals that the initial three topics undergoing more extensive study over the course of time compared to the other topics (the volume ratio in the period increased more than the other topics as time progressed) were Topic_1, Topic_3, and Topic_2 (“Augmented Reality in Education and Cultural Heritage”, “Enhancing Safety and Information in Food Consumption”, and “Medical Education and Patient Care”). These topics are also the first three topics in the “high volume/high velocity topic” group. The first three topics that have decreased in volume over time (less studied compared to other topics) are Topic_8, Topic_5, and Topic_6 (“Laparoscopic Training and Stroke Rehabilitation”, “Augmented Reality in Electrical and Electronic Systems”, and “Gesture-Based Instruction and Maintenance”). Among these topics, Topic_8 (“Laparoscopic Training and Stroke Rehabilitation”) is in the “low volume topic/slow topic” category, while the other two (Topic_5 and Topic_6) are in the “medium volume topic/fast topic” category. With the help of the Acc values in Table 5, it is visualized in Fig. 8 which topics are studied more and which topics are studied less over time among all topics.

Fig. 8

Sequential study intensity of topics compared to other topics in periods

In Fig. 8, it can be seen that the study intensity of eight out of the eleven topics increased over time, while the intensity of the remaining three topics decreased. The first three topics experienced a rapid increase in study intensity compared to the others, while the remaining five topics had a relatively slower increase. Among the three topics that experienced a decrease in study intensity, the first two decreased at a slower rate, while the last topic decreased rapidly. This indicates that the study intensity of the last topic decreased quickly. Finally, the top five subjects with the highest volume in each period were ranked, revealing the top five most studied subjects in each period compared to the other topics. The results of this ranking are visualized in Fig. 9.

Fig. 9

Top five topics that stand out compared to other topics according to the intensity of study in each period

As can be seen in Fig. 9, while the topics in the first two periods are more variable (seven different topics including Topic_1, Topic_2, Topic_3, Topic_4, Topic_5, Topic_6, and Topic_8), the topics in the last three periods, although the order varies within itself, have emerged as the top five topics ordered by volume. As a matter of fact, the top five topics in the last period were obtained directly as the top five topics ordered by volume.

4 Discussion

4.1 Discussion on year, authorship, geographic trends, and journal impact

This section discusses the findings in the context of the first two research questions (RQ1 and RQ2.) According to the findings, educational AR studies between 2003 and 2022 showed a very rapid increase after 2015. It is seen that technological advances in mobile devices and software and hardware released by various companies recently contributed to this increase (Arth et al., 2015). At the same time, the increase in users and companies interested in this technology shows that the rise in market share has attracted the attention of journals and scientists in health (Denche-Zamorano et al., 2023), entertainment, and other fields. According to a study examining the period between 2017 and 2021, the number of citations on this subject is higher than the number of publications, indicating that the demand for high-quality AR publications has increased (Jajic et al., 2022). In light of this information, reasons such as technological developments after 2015, the increase in demand for quality AR publications, and the rise in financial support from companies can be considered influential on the increase in publications in this field. The most prolific authors in this field are Billinghurst, M. (from Australia), Mantri, A. (from India), and Ong, S.K. (from Singapore). Billinghurst is cited as the most prolific and also the most cited author for conference papers, books, and book chapters (Avila-Garzon et al., 2021). It seems that Billinghurst has been working on AR since 2001. Therefore, it is natural that he is the most prolific writer. Mantri has been working on AR since 2015, but recent years have been the time when he has been most productive in this regard. Ong has been working on this subject since 2004.

According to publication origin, the United States, China, and Spain are the countries with the most publications in this field, respectively. The USA and Spain are the countries that contributed to this field in 2014 and before (Avila-Garzon et al., 2021). It is seen that China has contributed rapidly to this field in the recent period and has risen to second place. Since the USA and Spain are countries that have been publishing in this field for many years, it is natural to encounter this result. China, on the other hand, has made a move in the field of educational AR in recent years, as in every field of technology, and has taken its place in the top three rankings.

According to the findings, “Applied Sciences” is one of the journals where studies in this field are published the most. Some of the bibliometric analyses have determined that the journal with the highest number of publications is “Computer and Education” (Aslanci & Bayrak, 2023; Avila-Garzon et al., 2021). From this point of view, it can be said that the journal “Applied Sciences” has given more space to educational AR articles in recent years. In addition, the fact that the journal “Applied Sciences” is open access may mean that it has become more preferred by authors. The results of the study by Denche-Zamorano et al. (2023) support this idea.

4.2 Discussion on the evolving landscape of educational AR topics

The analysis indicated that educational AR studies fell into eleven distinct topics. It was noted that the sequence of volume ratios and acceleration values for the emerging topics was nearly identical. The top three most voluminous topics were “Augmented Reality in Education and Cultural Heritage”, “Medical Education and Patient Care”, and “Enhancing Safety and Information in Food Consumption”. In addition, these topics have been studied more than other topics over time (see Table 6) and have the highest study intensity compared to other topics (see Fig. 8). It is also seen that they are among the most studied current topics (see Fig. 9). AR technology offers important opportunities in cultural heritage education. AR proves to be an optimal technology for reconstructing historical events while maintaining the authenticity of original architecture and landscapes (tom Dieck & Jung, 2017). Increasing attention is being directed towards AR applications, spanning from 3D reconstructions to intangible cultural heritage. AR offers a significant potential to engage the younger generation in an educational sense and to fully engage them in historical environments (Boboc et al., 2022). When educational AR research is compared with AR applications in other fields, it is seen that this technology offers an entertaining environment that is especially suitable for the entertainment understanding of young generations (Capecchi et al., 2024). On the other hand, AR applications play an important role in the fields of entertainment and cultural richness by enriching users’ experiences and providing interactive and entertaining content (Cranmer et al., 2023). From these perspectives, it is predicted that AR technology will continue to bring cultural heritage to life for years to come. Similarly, in medical education, students have the chance to experience the course content without risking patient safety. AR applications have been developed in many subjects in medical education, especially anatomy. It has been stated that students generally give positive feedback on these applications and that they enrich the learning experience (Dhar et al., 2021). The fact that the main subject of medical education is the human body naturally brings ethical and safety issues to the agenda. “Medical Education and Patient Care” is also the second subject with the highest number of publications. There is AR software used in the field of healthcare for non-educational purposes such as surgical procedures and visualization of information (Yoo et al., 2024). These programs are important in terms of facilitating all health-related processes (Moldovanu, 2024). Therefore, it can be safely said that medical education will continue to benefit from AR technology for many years to come.

Increasing awareness of food consumption has become especially important recently. Knowing which type of food (carbohydrate, protein, etc.) you consume and having information about foods that should not be consumed too much, especially carbohydrates, are crucial for the quality of life. Especially in diseases such as diabetes, it is vital for people to consider what they eat (Juan et al., 2019). In this context, AR is very successful in increasing nutritional awareness (Paramita, Yulia, & Nikmawati, 2021). The important role of eating quality in preventive health and the expectation for an increase in effective AR practices in nutrition education (Paramita et al., 2021) indicate that studies in this field will continue to gain momentum. The top three most voluminous topics are also the fastest-accelerating topics. This means that studies on related topics are increasing rapidly. These are fundamental topics in life whose importance and popularity are not time-dependent.

In volume terms, the topics with the lowest volume and the slowest pace were “Serious Games for Children with Autism Spectrum Disorder”, “Augmented Reality in Chemistry and Biology Laboratories”, and “Augmented Reality for Safe and Efficient Driving,” respectively. These topics exhibit lower research intensity compared to others. Notably, research utilizing AR technology with individuals diagnosed with autism spectrum disorder (ASD) has yielded promising outcomes in areas such as social interaction, emotion recognition, attention skills, and functional learning (Berenguer et al., 2020). However, since AR technology is a relatively new technique, high-quality empirical studies for diseases such as ASD may be scarce. This suggests that this topic is not one of the topics with decreasing study intensity but rather ranks lower among the topics with increasing study intensity (see Fig. 8). Accordingly, it is strongly predicted that more studies will be conducted on this topic in the future. It is noteworthy that AR technology stands out as an effective educational tool in chemistry education and can be used as a facilitator in various subjects (Mazzuco et al., 2022). There are inconsistencies regarding the effectiveness of AR technology in higher education. While a two-dimensional representation can be used for all levels of education in chemistry courses, AR technology was found to be more useful for younger students (Fombona-Pascual et al., 2022). Accordingly, it is seen that more studies are needed for chemistry education. It is seen that useful results have also been obtained in biology teaching with AR technology (Weng et al., 2020). However, the scarcity of studies, especially in the field of biology, draws attention. However, the subject with the highest number of publications over time (see Table 5) is “Augmented Reality in Chemistry and Biology Laboratories”. Hence, there is a crucial need to boost the quantity of comprehensive experimental investigations concerning AR within the realm of science education, encompassing physics, chemistry, and biology. This step is essential to thoroughly showcase the efficacy of this technology, and it is evident that research in this domain is poised to grow. Notably, the topic of safe driving emerges as the least voluminous and least accelerated area of study. However, according to the time periods in which these studies were conducted, it is clear that it is a topic that has recently gained momentum. It is known that AR, which is a useful tool in science education, is also used in the field of engineering for purposes such as system maintenance and the assembly of complex systems (Solmaz & Van Gerven, 2022). It is clear that science and engineering fields will continue to benefit from the powerful opportunities offered by AR.

While studies on driving safety with AR technology emphasize the benefits of this technology, they point to some problems that need to be overcome, such as the difficulties that arise with the driver’s attention and visual perception (Boboc, Gîrbacia, & Butilă, 2020). It is therefore clear that driving-related AR technologies need further research. Considering that it is a newer subject of study compared to other subjects and that the technological competencies required by AR are beginning to be met more and more today, it is expected that studies on driving will increase in the coming years. It is known that AR software related to driving techniques is also used for gaming purposes in the entertainment industry.

When we look at the time periods in which the topics were mainly studied, it is seen that “Augmented Reality in Electrical and Electronic Systems” and “Gesture-Based Instruction and Maintenance” were studied in the early years of AR technology. Considering that this is the development stage of AR technology, it is natural that related topics stand out. These topics are among those whose study rates have decreased over time. Accordingly, it can be said that the related topics are no longer current. When the time periods are analyzed, it is seen that both the developments and improvements in AR technology and the issues that emerged with the differentiation of the interests of the changing social structure come to the fore. For example, it is seen that AR technology started to be utilized in the early 2010s (Ma, Jain, & Anderson, 2014), when social awareness of healthy nutrition was trying to be increased. However, this was made possible by improved and accessible AR technology. Since 2015, with the widespread adoption of mobile phones and the advancements in their technological capabilities, the variety of areas using AR technology has also increased. In this period, applications in the category of “Serious Games” were developed and started to be used.

To summarize, the topic modeling analysis reveals a dynamic and evolving landscape within educational AR research. The prominence of topics like “Augmented Reality in Education and Cultural Heritage”, “Medical Education and Patient Care,” and “Enhancing Safety and Information in Food Consumption” underscores their enduring significance and likely continued growth. These areas offer rich opportunities for exploring how AR can enhance learning experiences, promote accessibility, and address contemporary societal needs. The observed acceleration in studies related to these topics suggests a growing awareness of AR's potential and a need to refine research methodologies and address emerging ethical considerations. Furthermore, the emergence of novel topics such as “Serious Games for Children with Autism Spectrum Disorder” and “Augmented Reality for Safe and Efficient Driving” highlights the broadening scope of AR applications in education, and its potential for addressing diverse learning needs. Understanding the historical trajectory of topics like "Augmented Reality in Electrical and Electronic Systems" and "Gesture-Based Instruction and Maintenance" reminds us that the field is constantly adapting to technological advancements and evolving societal interests.

4.3 Implications and insight for the near future

The results of this study not only reveal the current state of the field through semantic content analysis but also show various trends that may shape the near future of this field. One of the important outcomes of the study is to see the area in which the field will evolve in the near future with these trends and to make inferences for researchers and practitioners in the field. The prominence of topics such as “Augmented Reality in Education and Cultural Heritage”, “Medical Education and Patient Care,” and “Enhancing Safety and Information in Food Consumption” underscores their sustained importance and likely continued growth. The observed acceleration in studies related to these topics implies an increasing research focus, potentially fueled by societal demands for innovative educational approaches and the technological capabilities of AR. The identified emerging areas, including “Serious Games for Children with Autism Spectrum Disorder” and “Augmented Reality for Safe and Efficient Driving,” are poised to gain significance in the near future, reflecting a growing awareness of diverse applications for AR in specialized educational domains. Additionally, the historical trajectory of topics like “Augmented Reality in Electrical and Electronic Systems” and “Gesture-Based Instruction and Maintenance” indicates a shift in research emphasis over time, suggesting that ongoing technological advancements and evolving societal interests will continue to shape the landscape of educational AR research. Overall, the findings imply a dynamic and multifaceted future for AR in education, driven by both the sustained importance of established topics and the emergence of novel, specialized applications.

This research offers valuable insights for educators and practitioners seeking to integrate AR into their teaching practices. The prevalence of studies on cultural heritage, medical education, and food safety suggests these areas are ripe for AR-based interventions. Educators should consider leveraging the interactive and engaging nature of AR to enhance student understanding and engagement in these domains. Additionally, the emergence of topics like "Serious Games for Children with Autism Spectrum Disorder" and "Augmented Reality for Safe and Efficient Driving" signifies the growing relevance of AR in specialized learning environments. Practioners should explore the potential of AR for supporting students with diverse learning needs and addressing contemporary challenges such as driving safety. As the field of educational AR continues to evolve, practitioners should stay informed about emerging technologies and research findings to ensure that their AR interventions are both effective and ethical.

5 Conclusions

In conclusion, this comprehensive analysis sheds light on the evolving landscape of educational AR literature over the past two decades (2003–2022). The descriptive analysis reveals a significant surge in publications, particularly after 2015, highlighting the growing interest and exploration in AR applications for educational purposes. Prolific authors such as Billinghurst, M., Mantri, A., and Ong, S.K., along with leading nations like the United States, China, and Spain, dominate the academic discourse in this field. Top journals, including “Applied Sciences” and “Computers and Education,” showcase the diverse publishing landscape.

The topic modeling analysis identifies eleven primary topics. The top three most voluminous topics were “Augmented Reality in Education and Cultural Heritage”, “Medical Education and Patient Care”, and “Enhancing Safety and Information in Food Consumption”. In addition, these topics have been studied more than other topics over time and have the highest study intensity compared to other topics. In terms of volume, the least voluminous and slowest topics have been “Serious Games for Children with Autism Spectrum Disorder”, “Augmented Reality in Chemistry and Biology Laboratories”, and “Augmented Reality for Safe and Efficient Driving”, respectively. When we look at the time periods when the topics were mainly studied, it is seen that “Augmented Reality in Electrical and Electronic Systems” and “Gesture-Based Instruction and Maintenance” were studied in the early days of AR technology. Since 2015, applications in the “Serious Games” category have been developed and started to be used. The five most studied current topics are “Augmented Reality in Education and Cultural Heritage”, “Medical Education and Patient Care”, “Enhancing Safety and Information in Food Consumption”, “Enhancing Learning Experience”, and “Augmented Reality in Electrical and Electronic Systems”. This study serves as a valuable resource for researchers, educators, and policymakers navigating the multifaceted realm of educational AR, offering insights into current trends, emerging topics, and areas warranting further exploration in the future.

6 Limitations and future works

This study aims to examine the trends and prevalence of educational AR studies covering the 2003–2022 period. This innovative and unique research employs topic modeling analysis to investigate the application of AR technology in the field of education. The findings are anticipated to provide insights for future research, uncovering both overt and covert patterns within the literature on educational AR studies. However, the study has some limitations. First of all, the study was limited to analyzing only articles. Future research studies may consider including all types of documents including conference proceedings, book chapters, etc. in the corpus. Another limitation is related to the nature of the LDA algorithm. Although LDA-based topic modeling offers a method based on real parameters, it has some limitations due to limitations in the number of predefined topics and potential overlap. This can create difficulties in fully capturing dynamic concepts and fully analyzing complex concepts. In future work, improvements in the area of LDA-based topic modeling can be proposed, such as flexible parameter settings, dynamic topic counting, and more efficient handling of overlapping topics. Furthermore, the exploration of new methods for more effective modeling of word order and semantic relationships within documents could be an important focus for future research in this area. Future studies may provide a more nuanced understanding of temporal evolution by providing a more detailed breakdown of trends over shorter time periods, focusing particularly on the increase after 2015. In this way, it can enable a more detailed analysis of changes over time and a more in-depth understanding of important developments, especially in the period after 2015. On the other hand, further research can be conducted using temporal analysis to confirm the results of this study. This will enable monitoring of future changes in increasing or decreasing trends in topics and allow for a better understanding of how the situation observed in this study may evolve over time.

Appendices

Appendix-A

Topics, the first fifteen terms that make up the topics, and volume ratios

Topics	The first fifteen terms representing the topics	Volume Ratio
Topic_1	technology, learning, design, application, educational, education, study, mobile, development, heritage, process, experience, analysis, cultural, museum	27.22%
Topic_2	medical, anatomy, health, education, virtual, care, clinical, technology, medicine, dental, teaching, patient, covid, health, care, pandemic	15.87%
Topic_3	food, mobile, library, game, commerce, estimation, safety, device, carbohydrate, service, information, entrepreneurship, technology, iphone, consumption	13.04%
Topic_4	learning, student, education, technology, study, learner, science, motivation, teaching, course, classroom, environment, engineering, experimental, experience	11.60%
Topic_5	circuit, image, ultrasound, video, electronics, pen, spherical, force, system, handwriting, machining, electrical, application, icp, physic	11.50%
Topic_6	system, assembly, user, object, real, image, instruction, tracking, maintenance, task, information, interaction, gesture, robot, time	7.06%
Topic_7	surgery, surgical, navigation, surgeon, planning, patient, robotic, intraoperative, neurosurgery, preoperative, procedure, accuracy, model, imaging, placement	6.45%
Topic_8	training, simulator, simulation, laparoscopic, skill, participant, surgical, task, rehabilitation, patient, trainee, performance, time, stroke, procedure	2.94%
Topic_9	child, game, study, autism, disability, intervention, learning, disorder, medium, asd, student, social, vocabulary, serious, school,	1.61%
Topic_10	laboratory, chemistry, chemical, lab, molecular, visualization, sandbox, student, structure, molecule, undergraduate, biology, topographic, biochemistry, climate	1.22%
Topic_11	system, driver, user, driving, computation, cue, vehicle, interface, bci, welding, reinforcement, network, visual, evacuation, machine	0.82%

1. Topic_1: Augmented Reality in Education and Cultural Heritage
2. Topic_2: Medical Education and Patient Care
3. Topic_3: Enhancing Safety and Information in Food Consumption
4. Topic_4: Enhancing Learning Experience
5. Topic_5: Augmented Reality in Electrical and Electronic Systems
6. Topic_6: Gesture-Based Instruction and Maintenance
7. Topic_7: Enhance Surgical Precision and Patient Safety
8. Topic_8: Laparoscopic Training and Stroke Rehabilitation
9. Topic_9: Serious Games for Children with Autism Spectrum Disorder
10. Topic_10: Augmented Reality in Chemistry and Biology Laboratories
11. Topic_11: Augmented Reality for Safe and Efficient Driving

Appendix-B

Matrix showing the distribution of the number of publications of the topics over the years and the acceleration values of the topics

	Years
Topics	2003	2004	2005	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022	Total	Acc
Topic_1	1	1	1	0	5	4	6	4	7	17	21	42	39	48	61	76	125	144	195	219	1016	9.84
Topic_2	1	1	1	2	2	2	1	2	8	5	11	17	14	25	37	36	72	114	180	230	761	8.23
Topic_3	0	1	0	0	2	1	4	7	14	10	22	15	26	19	47	61	68	83	95	139	614	5.75
Topic_4	0	1	1	2	1	1	4	1	4	9	11	22	14	22	32	40	52	81	78	99	475	4.51
Topic_5	1	1	1	2	3	3	5	4	7	8	10	11	6	13	20	29	42	56	65	58	345	3.03
Topic_6	2	2	4	2	4	2	8	5	9	3	15	5	16	11	19	16	27	34	38	47	269	1.96
Topic_7	0	0	0	0	1	0	0	1	3	1	3	3	3	3	4	5	8	14	11	18	78	0.72
Topic_8	0	0	1	0	0	4	3	5	0	3	7	1	2	6	3	2	9	7	9	11	73	0.45
Topic_9	0	0	0	0	0	0	0	0	0	1	0	0	0	2	3	6	4	6	5	15	42	0.46
Topic_10	0	0	0	0	0	1	0	0	0	0	0	0	1	1	1	0	0	14	4	4	26	0.28
Topic_11	0	0	0	0	0	0	0	0	0	1	0	2	0	0	2	1	1	2	4	6	19	0.19
Total	5	7	9	8	18	18	31	29	52	58	100	118	121	150	229	272	408	555	684	846	3718