Because of the great potential of AR technologies among museums worldwide, our study reviewed what has been researched in the existing literature corpus that explores privacy issues related to AR applications by museums to engage museumgoers better. The literature review method has become a popular research approach for digital reality technology researchers in recent years. For example, Guzman et al. (2020) examined the privacy and security impacts of AR and MR technologies, and they claimed their study to be the first study dealing with these critical issues. Similarly, Song et al. (2021) also adopted a literature review method to study the role of AR in digital fabrication. The literature review method is also famous for analyzing existing cross-cultural psychology literature (Obschonka et al., 2022).

Our study employed a literature review approach to situate our study within the cross-cultural psychology perspective and focus on a thorough examination of existing literature to investigate how scholars have studied issues related to how museum consumers deal with these privacy-invasive AR applications. Our approach follows the literature review research conventions to describe and analyze emerging research trends described in our RQ1. However, unlike existing literature in the field that relies on the subjective categorization of research and the articles retrieved from keyword searches (Guzman et al., 2020; Song et al., 2021), we employed a text mining method to extract emerging keywords and key phrases related to the applications of AR technologies by museums from a large number of research articles. Although we have stated our research questions as RQ1 and RQ2, they are related because extracted keywords and key phrases in RQ1 will contribute to our analyses for RQ2.

To summarize, our study aims to answer the following research questions:

The literature review method has been popular in digital reality technology research (Guzman et al., 2020) and has been increasingly adopted by cross-cultural psychology researchers (Obschonka et al., 2022). While these early literature review studies focus on the categorization of the technologies and their related accessories, such as head-mounted devices, the “value-sensitive approach” has emerged in 2000 (Friedman and Kahn, 2003) to investigate data ownership, integrity, privacy, and secrecy issues (refer to Guzman et al., 2020, p. 110:3–110:4 for a complete review). After an extensive keyword search of existing popular databases, our study is the first study to use a text mining study to explore how the existing literature is studying potential privacy infringement as one of the most noticeable social impacts of AR technologies by museums (Franziska et al., 2014; Kang and Yang, 2020).

As a recently emerged method for scholars to study a variety of phenomena with a minimum amount of human interference (Lin et al., 2016; Yang and Kang, 2018; Kang and Yang, 2022), this objective data categorization and interpretation approach can address problems that we have found in traditional literature review analysis and meta-analysis studies (Guzman et al., 2020; Song et al., 2021). Additionally, a text mining method also offers researchers the excellent opportunity to systematically review and analyze many retrieved articles objectively, without human errors (Lin, cited in Yang and Kang, 2018). For example, Song et al. (2021) analyzed only 84 articles. Our study, on the contrary, studied 715 articles from keyword searches. In the cross-cultural psychology field, Obschonka et al. (2022) studied merely 79 articles collected in 2020. Therefore, we argue that the extent and the scope of our literature review method are comparable to existing studies.

This text-mining literature analysis study used QDA Miner and WORDStat to analyze emerging keywords and key phrases in the literature corpus. QDA Miner can easily analyze a large number of documents “in a variety of file formats including M.S. Word, WordPerfect, RTF, PDF, HTML, XML, MS Access, Excel, SPSS, Paradox, dBase, QSR N6, Nvivo, Atlas.ti, HyperResearch, Ethnograph, Transana, Transcriber” (LaPan, 2013: 775).

We have chosen these two databases to compile our literature corpus because of their comprehensive coverage of business- and engineering-related topics. Business Source Complete covers research areas in accounting, advertising, banking, management, and marketing, while Engineering Village (E.I.) includes over 190 subject areas in engineering research. We used the keyword phrases [“augmented reality” and “privacy” and “museums”] to search these two databases. The results have generated zero articles in both databases, suggesting the innovativeness of our research topic. To refine our searches, we revised our key phrases to [“augmented reality” and “privacy”] (1st keyphrase pair) and [“augmented reality” and “museums”] (2nd keyphrase pair). We generated 44 articles from Business Source Complete from the first pair and 21 from the 2nd pair. Using Engineering Village (E.I.) database, we generated 237 articles from the 1st pair, while 413 articles were retrieved from the 2nd pair. We compiled a literature review corpus of 715 articles from 1996 to 2022, covering peer-reviewed journal publications, conference papers, conference proceedings, etc. We used the abstracts of these articles because they summarize the essence of each article.

Based on the text mining technique, we analyzed recurrent keywords and critical phrases, following the procedures by Miner (2012) and Kang and Yang (2022). Additionally, the algorithms in the QDA Miner include K-Means Clustering for text categorization and text-to-numeric transformation and Latent Dirichlet Allocation (LDA) for topic modeling (Valcheva, n.d.) that are most relevant to this study. For example, we first conducted the document-clustering algorithm to transform unstructured textual data in each retrieved abstract into a structured representation before applying this procedure to reduce the vector space through either direct clustering or dimensionality reduction (Miner, 2012).

We conducted several text mining techniques to provide empirical data to answer our first question. Our second research question aims to identify recurrent keywords and key phrases from the retrieved literature review corpus that studies the AR applications in the museum sector to demonstrate the research trends in the related topics. These techniques include the extraction of keywords and key phrases to estimate and demonstrate their relative importance among a-bag-of-words model in a Natural Language Processing algorithm in the corpus by examining Term-Frequency (T.F.) or TF-IDF (Term-Frequency-Inverse document Frequency) that shows the frequency of these terms (Teso et al., 2018; Yang and Kang, 2018; Kang and Yang, 2022). T.F. or TF-IDF statistics help researchers identify reoccurring words or phrases viewed as more important and prominent (Kang and Yang, 2022). A visualization of the word frequency statistics is a Word cloud analysis technique that has been a widely used text mining technique to demonstrate the frequency of keywords and key phrases in a graphical manner (Srivastava, 2014; Kang and Yang, 2022). As seen in Table 1 below, several keywords related to AR applications by museums have emerged; these include “privacy” (TF-IDF = 226.1), “security” (TF-IDF = 169.2), “data” (TF-IDF = 194.1), “information” (TF-IDF = 184.9), and location (TF-IDF = 157.0) (Refer to Table 1). Also noteworthy is the reoccurrence of consumer-related keywords, such as “user” (TF-IDF = 201.9), “users” (TF-IDF = 191.7), “visitors” (TF-IDF = 186.7), “experience” (TF-IDF = 171.5), “experiences” (TF-IDF = 112.8), and “visitor” (TF-IDF = 102.0) (Refer to Table 1).

The key phrase extraction procedure in QDA Miner and WordStat is to extract the most salient phrases in the literature corpus useful for “document categorization, clustering, indexing, search, and summarization” (Kang and Yang, 2020). As seen in Table 2 below, the key phrases related to the study of consumers’ concerns over privacy-related issues about AR applications by museums, “security and privacy” (TF-IDF = 169.5) is among the most prominent phrases in the compiled corpus, followed by “privacy concerns” (TF-IDF = 61.7), “privacy protection” (TF-IDF = 48.2), “location privacy” (TF-IDF = 46.8), “privacy and security” (TF-IDF = 25.2), “privacy risks” (TF-IDF = 24.4), “user privacy” (TF-IDF = 21.8), and “privacy preserving” (TF-IDF = 22.1) (Refer to Table 2). Our study’s significant findings demonstrate the scope, extent, and topology of consumers’ privacy concerns regarding museums’ AR applications in the literature. For example, while it is anticipated that “privacy concerns” emerge as a repetitive key phrase, this study also observes that more specific privacy concerns have emerged, ranging from “location privacy,” “user privacy,” and “privacy and security.” Additionally, the phrase “privacy preserving” has emerged to demonstrate how engineers investigate the “privacy by design” principle to ensure consumer privacy protection.

Another significant finding from our text mining research is to investigate further the privacy concerns related to the utilization of consumers’ location privacy information in offering AR museum-going experiences by interacting with AR-superimposed exhibitions or on-location museum guidance. Our study has also discovered another remarkable group of key phrases in the corpus about the location information related to AR applications in the museum sector. For example, “location-based services” (TF-IDF = 34.2), “personal information” (TF-IDF = 22.8), “location information” (TF-IDF = 19.0), and “sensitive information” (TF-IDF = 19.5). Additionally, the emphasis on how museum users or visitors would respond to these concerns has emerged as a major research topic. The terms “location-based services,” “location information,” and “personal information” have emerged. Additionally, it is interesting to note the relationship between location information and “sensitive information.”

Furthermore, despite the privacy concerns of these AR applications in the museum sector, the literature corpus has demonstrated that museum curators and professionals attempt to make the best use of AR technologies to generate more positive museum experiences. Reoccurring key phrases include “user experience” (TF-IDF = 46.4), “visitors experience” (TF-IDF = 26.1), “visitor experience” (TF-IDF = 24.4), “immersive experiences” (TF-IDF = 20.8), and “user interaction” (TF-IDF = 19.0) (Refer to Table 2). The TF-IDF statistics quantify the importance of a key phrase in the overall document corpus (Silge and Robinson, 2019, 2022). The statistics demonstrate privacy-related issues consumers perceive in the museum sector studied in the existing literature.

Trend analysis has been used along with conventional literature review to assess the journal metrics (Kokol, 2017) and to compare methodologies used in academic studies over a long period and across geographical regions (Umer et al., 2018). Unlike the human coding of collected 153 service marketing articles (Umer et al., 2018), our study demonstrates that a text mining literature review can objectively analyze many articles (N = 715). While Umer et al. (2018) focused on methodology, geographic regions, and years of publications, one of the significant findings from our study is that we have examined research themes and variations between 1996 and 2022 and have visually presented that the emergence of key research topics from our literature corpus. For example, extracted key phrases that are most pertinent to our study include “security and privacy,” “privacy and security,” “privacy risks,” “privacy concerns,” “privacy issues,” “user privacy,” “location privacy,” “privacy protection,” and “privacy preserving” (Refer to Figure 1 below). The visualization of the numerical data contributes to our understanding that privacy-related issues have emerged significantly since 2010, with their highest growth after 2016.

The following year-by-year analyses of selected key phrases present another significant finding of our study to allow researchers to comprehend the research trends as demonstrated in the existing literature. It is noteworthy that, while research on “privacy issues” first appeared in 1996, followed by sporadic research on “privacy and security” (2002), “security and privacy” (2006, 2007), the exponential growth of the privacy-related literature can be found after 2015 when the number of related articles jumped to 12. Privacy issues related to AR applications have since been gaining interest among scholars, with the total publications reaching 11 in 2016, 17 in 2027, 11 in 2018, 25 in 2019, 20 in 2020, and 18 in 2021. The rapid increase of these publications demonstrates AR technologies on consumer privacy. Extracted key phrases related to how consumers would respond to the implementation of AR technologies by museums and its impacts on users’ experiences include “user experience,” “user interaction,” “visitors experience,” and “visitor experience” (Refer to Figure 2 below). While research on “visitor experience” first appeared in 2005, scholarly interest in user-related topics increased in 2014 with 6 articles, followed by 5 in 2015, 6 in 2017, 8 in 2019, 13 in 2020, and 8 in 2021. In particular, research on visitor experience has increased from one article in 2015, 2 in 2013, 4 in 2018, 5 in 2019, 4 in 2020, and 5 in 2021. The increase of these publications demonstrates that AR technologies by museums have been related to their impacts on visitor experiences. The demonstration of research topic variations offers museum scholars and practitioners a road map to develop their research and application strategies when integrating these innovative AR technologies.

As demonstrated in this text mining study of privacy literature related to the implementations of AR in the museum sector, our literature review study has confirmed the growing importance of these topics in the past 5 years. Our study summarizes what research topics have been gaining attention among scholars. The increase in publications after 2015 has collided with the rapid deployment of AR technologies. With the heightened security and privacy concerns among consumers, new research topics such as “privacy preserving” in system design emerged in 2014, along with more research on “privacy protection.” Additionally, the growing focus on visitor(s) experience has echoed the implementation of AR technologies to enhance museum visitors’ experiences inside the museums. However, the unintended and unanticipated negative impacts have infringed on consumers’ privacy. This text mining study of existing literature on these essential topics has offered scholars a longitudinal analysis of trends in research topics. Unfortunately, digressing from our original plan to locate and analyze cross-cultural psychological research on consumers’ perceptions and strategies to address location privacy invasion due to AR applications in museums, our study shows an ample opportunity for psychological scholars to study this global phenomenon.

Like any literature review study, several limitations need to be considered to interpret better and understand our research findings. First, the thoroughness of sampled articles in our literature review corpus is a primary concern in any text mining research (Yang and Kang, 2018; Kang and Yang, 2019a, 2021b, 2022). This study only retrieved articles from the widely used social scientific Business Source Complete and technology-oriented Engineering Village (E.I.) databases. Despite our intention to be comprehensive, future scholars may generate more comprehensive findings by examining other databases. Second, another limitation is processing words, keywords, key phrases, and lexicon dictionaries to extract recurrent linguistic patterns and trends through a bag-of-words approach (Teso et al., 2018; Kang and Yang, 2021b). Some text-mining scholars have claimed that the heavy dependence on a single word may fail to address the diversity of word meanings (i.e., polysemy) (Teso et al., 2018; Kang and Yang, 2021b).

Additionally, the lack of contextual information related to each keyword and key phrase may not probably address the relative importance of each research topic (Teso et al., 2018; Yang and Kang, 2018; Kang, 2019; Kang and Yang, 2021b). Scholars may explore the “keyword in context” function in QDA Miner can benefit their research. Lastly, this study only collected and analyzed studies published in the English language and archived in the selected commercial databases. Studies published outside of these linguistic and database parameters were not text mined. Future scholars from other language regions may benefit from analyzing non-English academic research to shed light on these issues.