1 Introduction

Landscape elements are essential in tourism and recreation. Several studies have attempted to develop a more comprehensive approach to understanding the interrelationship between tourism and the satisfaction experienced in a landscape (Jiménez-García et al., 2020; Mikulec & Antoušková, 2011). This relationship is influenced by several factors, such as the type, scale, and location of the landscape. As one type of landscape, large-scale infrastructure could play an important role in tourism and have attracted the attention of tourism-related researchers (Reid et al., 2005).

When considering the experiences and perceptions that arise through landscape and tourism activities, several issues emerge that are applicable to specific fields of tourism (Stoffelen & Vanneste, 2015). For example, a tourism-landscape framework has been used to enhance the understanding of geo-tourism, and particular landscape features that impact visitors’ experiences have been identified in rural tourism. In addition to forms of “new tourism,” such as ecotourism and cultural heritage tourism, there is also increasing demand from the tourism industry to foster interest in large-scale engineering projects that use sophisticated technologies to build artificial structures, which reflect the social arguments underpinning such infrastructure projects.

The landscape aesthetics and related cultural ecosystem service (CES) content of these infrastructure projects have been previously discussed, including issues such as landscape changes arising from the planning and construction of projects such as roads (Marianne et al., 2017), new hydroelectric developments (Chen et al., 2018), and wind turbines (Frantál & Kunc, 2011). Although it has been indicated that a better understanding of the landscape from the perspective and framework of tourism is necessary (Smith & Ram, 2017), studies on the tourism and landscape experiences provided by large-scale infrastructures as tourist attractions are still few.

Accordingly, the purpose of this study is to explore tourists’ perceptions of landscapes concerning large infrastructure projects and assess how these perceptions change in relation to external factors. More precisely, the study’s questions are as follows: (1) How do tourists perceive the landscape values of infrastructure differently according to the different structural forms of projects (e.g., bridges vs. hydraulic dams)? (2) How do visitors’ landscape values vary based on their different cultural backgrounds? (3) How do visitors’ landscape values vary based on whether the same structures are placed in different locations and environmental conditions, including urban and/or well-developed tourism destinations?

The study is framed by a landscape cultural values model, which focuses on tourism in relation to large-scale structures. We considered the online travel reviews (OTRs) posted by tourists with different cultural backgrounds about four main visiting sites: three bridges and one dam project in Japan. We developed a four-step procedure for the content analysis of the reviews and analyzed the results using a quantitative statistical test.

2 Related research

2.1 Landscape values

Various approaches have been employed to investigate the human perception and visual assessment of landscapes. Using the two basic paradigms of objective and subjective viewpoints, four approaches to landscape evaluation can be identified: incorporating expert, psychophysical, cognitive, and experiential perspectives (Zube et al., 1982). In terms of design and planning practices, the expert paradigm has been the most used in visual landscape assessments. However, the public should also be involved in evaluating the value provided by certain places.

One comprehensive way of understanding the human perception of landscape is to examine the various values associated with people’s feelings, images, and sociocultural lifestyles. Through a synthesis of research on landscape values, Butler (2016) identified landscape-related values communicated through the specific assessment of categories, including economic value, natural significance, aesthetic/scenic value, recreational value, cultural significance, and intrinsic value.

Chen et al. (2018) undertook a spatial exploration of the landscape values embedded in assessments of hydroelectric dam projects, grou** value-related aspects of textual information and online photographic data into six types: aesthetics; sense of home; community attachment; lifestyle; memory; and cultural identity. The evaluation of landscape values is also linked to the more general context of perception assessment, such as ecosystem services. For example, Smith and Ram (2017) extracted four main aspects of value—spiritual, emotional, intellectual interactions, and existence—that work as landscape advantages for tourists and visitors. The identification of each item is somewhat abstract, and thus, it is usually difficult to apply it directly to various cases. Researchers have adapted the framework using their subjective judgment, and thus, the framework always remains somewhat discrepant between studies. A more generalized and applicable approach is sought to analyze in detail perceptions of the landscape as they relate to tourism.

Stephenson (2008) developed an integrated cultural values model (CVM) that conceptualizes the multiple ways in which people perceive value in different aspects of the landscape. Her approach integrates the two basic landscape concepts of biophysical and immaterial phenomena. Based on case studies in New Zealand, she identifies three types of values attached to landscapes: first, physical, tangible, and measurable aspects (e.g., physical land features, natural elements, and human-made structures); second, practices, which include human practices and natural processes (e.g., ecological processes and human activities); and third, relationships in which value is derived from people’s interactions with one another in the landscape or with the landscape itself (e.g., sense of place, aesthetics, sensory responses, memories, meanings, and ecological relationships). The values attached to the landscape are thus seen to contribute to cultural identities and environmental sustainability.

Several case studies have been undertaken using the CVM framework. Stephenson (2010) further developed the model by focusing on the involvement of temporal and spatial factors and described the CVM framework as interacting with space and time through either a static or dynamic approach to five main types of landscape qualities. The landscape model has been tested and adapted to different conditions. Bieling et al. (2014), for instance, interviewed both residents and visitors at various settings to explore the extent to which the landscape can affect people’s sense of well-being. Their results were reported to be consistent with the conceptual framework of Stephenson’s CVM while fitting some aspects of the ecosystem services framework.

Thompson-Fawcett and Picard (2015) applied the CVM approach to the Cardrona rural landscape, focusing primarily on its temporal dimension. Through interviews with both users and managers, specific landscape elements were clarified as belonging to forms, practices, and relationships; the model was then developed with preferences for future values as expressed by the informants.

Shaw and Bieling (2015) analyzed landscape values in areas of Germany, Australia, and Ireland using semistructured free-lists per an ethnographic approach. The results also verified the CVM framework, providing a rich understanding of the emergent interface of landscapes, mainly through a cultural lens. Therefore, the model is useful in integrating culture into sustainable landscape management and planning. The model was developed to understand the potential range of values that might be contained in a given landscape, to help address the problem of fragmented understandings of landscape value and to consider the contribution of the landscape to cultural sustainability. The model has further potential significance to be an integrated framework for landscape work with other disciplines by offering a shared frame of reference for landscape studies in those fields. In particular, it may be applied to both “ordinary” landscapes and highly valued landscapes, as in tourist sites, or to compare “outsider” views with “insider” views of the same landscape. Cross-cultural application of the model could also test how the framework can convey non-Western understandings of landscape (Stephenson, 2010).

In general, there are relatively few deep studies directly discussing large-scale infrastructure as tourism attractions from a landscape perspective. Research related to landscape value perception is mainly about general tourism destinations, such as a certain city or park, or general landscape forms (such as forests, rivers, lakes or plains), while the landscape values embedded in large-scale infrastructures remain unexplored, especially considering their emerging tourism function.

2.2 Use of user-generated content (UGC) and content analysis in tourism research

User-generated content, including online consumer reviews and blogs, is a valuable source of information in the fields of tourism and hospitality (Papathanassis & Knolle, 2011). The type of data in this category include texts (such as reviews), images, video, and live-streamed media. For most datasets and research themes in previous studies, content analysis proved to be the most frequently used procedure (Lu & Stepchenkova, 2015).

Previous studies have shown that UGC composed of a combination of text and image data generated by the research subjects themselves provides key insights into the tourist experience and perceptions of both landscape objects and tourism activities. One of the most popular applications of UGC is to measure how a particular tourism destination is experienced through the cognitive perceptions of visitors. For example, S. Choi et al. (2007) analyzed image representations of Macau on the internet by examining the content of a variety of web platforms and identifying different images of Macau projected by multiple information providers. Mak (2017) conducted a visual content analysis of photographs and a content analysis of textual data from both UGC and promotional material generated by national tourism organizations (NTOs). Through a comparative analysis of destination images of eastern Taiwan, differences between the two groups’ viewpoints were identified.

UGC can also be useful as data for exploring the tourist experiences at various sites. Pearce et al. (2015) attempted to categorize and interpret Chinese tourists’ online visual representations of an Australian landscape captured while traveling. For the field of landscape research, Tieskens et al. (2018) took social media photos from Flickr and Panoramio to estimate the correlation between landscape attributes and landscape preferences in a peri-urban area. Wartmann et al. (2018) extracted UGC information from Flickr images, tags and text from online blogs to investigate the relationships between the perceptions of landscape elements and sense of place, identified as CES items of visitors, at ten Swiss landscape sites. The application of UGC in this field provides great potential for evaluating the values perceived in landscapes for tourists and visitors.

Mass textual information is convenient for identifying the particular cognitive elements and experiences of subjects as text can be quantified for comparative analysis (Wartmann et al., 2018) or analysis in relation to other external factors. For instance, Lu and Stepchenkova (2012) extracted online descriptions of ecotourism experiences that US travelers had in Costa Rica by looking at 373 reviews from the TripAdvisor website. A comparative statistical analysis was undertaken based on variables that included 26 attributes identified by content analysis, together with the eco-tourists’ recorded levels of satisfaction with their journey. By comparison, images are usually used exclusively to reveal the visual perception of the landscape by using descriptive statistics for landscape elements. In some cases, information attached to the UGC data is also exploited, such as the extensive use of geo-tagging. Researchers combine geo-information with theoretical concepts relating to their research questions to explore spatial relationships and the distribution of data at specific sites (Hu et al., 2015).

Despite the widespread use of UGC, the analysis of this content is still limited to initial investigations in which features, such as the frequency of words, are compared or photographic images are taken as units of analysis (C. C. Lee & Hu, 2004; Stepchenkova & Zhan, 2013). Especially for factors regarding social constructs, a more scientific quantitative approach might be fruitfully integrated into theory-driven content analysis procedures.

3 Methodology

3.1 Overview

In total, 2173 tourist reviews on four infrastructure tourism (i.e. infra-tourism) sites in Japan were extracted as inputs for content analysis. These representative sites include three bridges and one dam. The text reviews were posted online on the TripAdvisor and Mafengwo websites by tourists with different cultural backgrounds. The texts were gathered by a web-scra** program developed for this study and were cleaned and then used as data in content analysis. Based on the CVM proposed in Stephenson (2010), a modified framework of culture-value categories was developed in this study to classify different aspects of landscape values at an infra-tourism site and build links between landscape value categories and corresponding tourist reviews. The segmentation operation was then conducted to extract single words from the review texts to establish a term list, which was matched and assigned to the categories for landscape value analysis. A term frequency–inverse document frequency (TF-IDF) index value was assigned to each word to evaluate the relative importance of a specific word in a document. Subsequently, the TF-IDF index value for each category was integrated to analyze the proportional distribution of terms across the categories. A comparison between two or more datasets was carried out for each category to examine statistical variations. The detailed methodology is shown in the subsequent sections.

3.2 Study sites

To examine the landscape values of typical large-scale infrastructures in various tourist perceptions, our study selected four sites that are considered good representations of large-scale infra-tourism sites in Japan (as shown in Fig. 1). They are Akashi Kaikyo Bridge, Kurobe Dam, the Irabu Bridge, and the Tokyo Rainbow Bridge.

Fig. 1
figure 1

Study sites of four main infrastructure tourism spots in Japan

Full size image

Completed in April 1998, the Akashi Bridge was built to be the world’s largest suspension bridge, lying 3,911 m long with a main tower that is approximately 300 m high. It is part of the Honshi Highway, one of the most important routes connecting the Honshu and Shikoku islands. To highlight the bridge structure, the landscape area was reorganized by the local government, and parks were planned for both the Akashi and Awaji sides of the bridge. From 2005 onwards, official commercial tours called the Akashi Kaikyo Bridge—Bridge World were offered to tourists from different countries. Based on existing tours of the Sydney Harbor Bridge in Australia, the Akashi official tour aimed to deepen public understanding of the business of road and bridge construction, thereby seeking to enhance public interest in the technology behind the social capital administration and management. Akashi Bridge was chosen as a target site to compare the perspectives of visiting groups from different cultural backgrounds and the effect of structural form on visitors’ experiences.

The Kurobe Dam, completed in 1963, is the largest and tallest arched dam in Japan. Its massive scale, superb surroundings, and impressive flow of 10 m3 of water per second have made it a popular tourist attraction, with more than 1 million annual tourists. It has become the most famous dam destination in Japan. Originally conceived as a hydroelectricity project, the design incorporated consistency and unity with the surrounding environment, representing the idea of environmental protection of biodiversity and water ecosystems in Kurobe. The dam is a prominent site of popularity in the dam-visiting wave, “dam-mania,” which has developed across Japan in recent years. Currently, the project is part of a tour package along with visits to surrounding destinations. The dam’s internal structure and water drainage systems are regularly open for viewing through such tours. The Kurobe Dam was selected as a target site to investigate the effect of structural form on visitors’ experiences, together with Akashi Bridge.

The Irabu and Tokyo Rainbow Bridges are also both large structures and famous destinations for visitors. The Irabu Bridge is located on Miyako Island, Okinawa, and is one of the most popular tourist destinations in Japan. Completed in 2015, the structure stretches 3,540 m, linking Miyako Island with Irabu Island as part of a prefecture road in Okinawa. The Tokyo Rainbow Bridge, officially called the Tokyo Bay Connector Bridge, is a 798-m suspension bridge that crosses from Shibaura Pier to the Odaiba waterfront in Minato Ward. The bridge was completed in 1993, and carries both the Capital Expressway and a subway route. Thanks to excellent scenic views of the skyline, especially at night, and its superb geographical location, it has become a significant tourist attraction, acquiring a symbolic status beyond its original conception. These two bridges, identified as two famous tourism spots in urban settings and typical tourism destinations, were used to investigate the effect of the environment around the same type of bridge structure on tourist perceptions.

3.3 Data sources and dataset construction

We chose the TripAdvisor website, which is the largest international online travel service, as a reliable primary source for collecting tourist-review data on the infrastructure in this study (Stockigt et al., 2018). Given that many foreign tourists are from China, another large online tourism website, Mafengwo, was also chosen as a source of review comments. We independently developed a web-scra** program to gather the review data for subsequent processing.

The standardized structure names were used as keywords for searching target review comments from 2008 to 2018. As the retrieved review texts are consistent with the language settings in the domain name, the data for each language group (Japanese, Chinese, and English) had to be extracted using the proper name in the corresponding language (the website search function can automatically provide this).

The variables in the raw dataset included reviewer ID, user ID, username, review publication date, review title, and review content. A preprocessing operation was performed after the data were gathered to clean the raw data. It should be noted that fake or spam reviews presented in these kinds of online platforms would have a significant impact on the validity of the analyzed results (Wu et al., 2020). For that reason, reviews with the following features were excluded from the dataset: (1) reviews that were too brief; (2) repetitive reviews (especially if the user was also a repeat poster), which could be due to problems with the website or the capture program; (3) reviews that were meaningless or of unknown significance, such as when the content contained only stock complimentary phrases; and (4) reviews that were advertorial in nature, such as those containing special symbols used for branding or promotional purposes. Moreover, we checked for users who post reviews at the same time. If the user profile has relatively rich information as well as historical travel records, the review is generally taken to be real and trustworthy, while if the user profile is mostly blank and the quality of the review content is doubtful, we chose to exclude it. Overall, however, the amount of data excluded was not significant.

The number of reviews in the prepared dataset ranges from 377 (Akashi Kaikyo Bridge) to 710 (Irabu Bridge). Within the review dataset for Akashi Kaikyo Bridge, the Japanese group has the largest review number compared to the Chinese group and English group. Chinese reviews of the Akashi site included 31 from TripAdvisor and 63 from Mafengwo, with more content posted on the former platform (as shown in Table 1). Unlike TripAdvisor, Mafengwo, which is mainly used within China, has a pure source of audience, and the commenting style is more likely to be influenced by the habits of domestic users. However, considering that the destination object itself is overseas (for Mafengwo users), which is consistent with the international nature of TripAdvisor, we combined the reviews from both platforms for word extraction and subsequent analysis.

Table 1 Overview of the datasets
Full size table

3.4 Categorization of landscape value types and word assignment

To comprehensively evaluate the landscape values of a large infrastructure site, we differentiate and categorize different aspects of the landscape that may be recognized and reviewed by tourists. Inspired by existing studies (Bieling et al., 2014; Stephenson, 2010), a new cultural value framework was developed by adding more subcategories, for example, the relationship items that evaluate landscape value based on human-landscape and human-site interactions. This system of categorization is believed to suit the specific case of large-scale infrastructures considered in our study. Table 2 shows the categorization system of landscape value types and the corresponding explanations and examples of typical words.

Table 2 Category system for review content based on the CVM framework
Full size table

Word segmentation was applied to the review content to extract single words as terms to be categorized. The segmentation process was performed using the KHCoder tool, which can deal with multilanguage texts (Murakami, 2018; Xu et al., 2019). A homemade dictionary and stop-words list were used in the segmentation processing, and the segmented results were saved as a term list (Velasco et al., 2017). After the term list was completed, the words extracted from the review content were assigned to different categories. The assignment process had to fully consider both the contextual meaning of each word and the operational definition of each category to ensure that specific units were correctly assigned. It was neither necessary nor practical to deal with all words in the term list; therefore, we set a threshold level for term frequency to ensure that only the more commonly used words were featured in our analysis.

Low-frequency words have little impact on representative approaches to textual content. As there was no consistent average number of reviews across the six datasets, we set the threshold at 10% of the highest frequency in the term list, rather than using a fixed number. This ensured that uncommon and unusual expressions occurring with relatively low frequency were filtered out and that the review content represented by the collection of segmented terms could be similarly isolated across different sites. In the end, approximately 200 words were included in each category, and the results were saved as term-category lists for each of the destinations.

To ensure the validity of the categorization, the coding process was conducted by the two authors separately; the results were then compared to examine any discrepancies between specific terms. In most cases, the precise meaning of the terms could be identified distinctly. Krippendorff’s alpha (Krippendorff, 2011) was used to characterize the agreement between the two authors on categorization, excluding the effects of “chance coincidence”. In general, the indicators satisfied the reliability requirement (see Table 3) (Hughes, 2021; Poldner et al., 2014). Difficulties might arise in the case of multiple meanings or specific contextual pairing, such as with the word “station,” which can relate to either site information or touristic activity. In these cases, the terms were placed back in their original context, and a judgment was made collectively regarding assignment to the most appropriate category.

Table 3 Reliability for each category group by destination
Full size table

3.5 Term and category scoring

To quantify the landscape value reflected from the review texts, we used a TF-IDF index for each word instead of a frequency index to calculate the importance of a specific word in a single document (Mee et al., 2021). Using the TF-IDF value as an evaluation weight in calculating the value distribution is a reasonable way to proceed, since it excludes the effect of word commonality and consequently reflects the actual importance and not just the numerical proportion of each category in the overall distribution (Mee et al., 2021; Menner et al., 2016; Pang et al., 2011). Generation of the content-category distribution was performed using an R tool, and the results were saved as content-category distributions that could then be statistically analyzed in the next stage. Through the representation process, the content of each review in unstructured, raw text was reorganized to fit the distribution of the theoretical categories of landscape values; the value of each category was obtained by collecting the TF-IDF values for each term in each category. Finally, the TF-IDF value of each landscape value category was transformed into a proportion to enable comparison across different objects or sites.

3.6 Statistical analysis

Based on the content-category distribution, a comparison of each category between two or more datasets was performed to find statistical variations. The datasets to be compared were grouped as one large set for analysis, with the variables identifying their type added according to the research questions. As the study primarily investigated the relationship between landscape value categories for each group, other variables were excluded, and only the categories relevant to the analysis were preserved as features of the review content records.

The proportional distribution of each category was seen as a continuous numerical value, while the group types for the datasets were set as classification variables. For the comparison between different group datasets, a nonparametric test, the Kruskal–Wallis test, was used because this test can handle the comparison of two or more multivalued groups. The results of the test can indicate if any variation in response values between the different groups can be considered statistically meaningful. Given the relatively large amount of review data in each dataset, this test enabled us to identify any significant differences within the content.

To visualize the diversity of the landscape value categories by different groups, we developed a differential tool initially derived from the structural topic modeling method (Roberts et al., 2013). A simulated distribution of the difference in the proportions of each landscape value category can be generated based on the comparison between two groups. If the distribution of differences is larger or smaller than zero, inclining to the side of one of the groups, the category would be perceived as more salient for the group that the distribution favored. When zero fell into a different scope, it was considered to indicate no significant difference in landscape value between the two groups (see Fig. 2). The Y-axis position of each topic is simply an arrangement of the topics and does not represent any other variable.

Fig. 2
figure 2

Differential tool derived from structural topic modeling

Full size image

4 Results

The results generally showed that the differences between the two groups in terms of review content were significant in most categories, suggesting that the perceived landscape values from tourists at any particular site can be notably influenced by the structural form of the site, e.g., a long-span bridge or an arched dam.

We conducted statistical tests on three comparison groups using the three research questions reported above. Using this approach, we tested the review content of the Akashi Bridge and Kurobe Dam for Question 1 (the effect of structural form on visitor experience); we tested the review contents for the Akashi Bridge from Japanese, Chinese, and English tourists for Question 2 (the perspectives of visiting groups from different cultural backgrounds); and we tested the review contents for the Akashi Bridge, Irabu Bridge, and Tokyo Rainbow Bridge for Question 3 (the effect of the environment applied to the same type of bridge structure).

The distribution of landscape values at each destination among the different factors is presented in boxplot charts and detailed data tables. We found that despite the similar overall distribution of the three types of landscape values, most of the categories for the subvalues showed significant differences (as seen in Fig. 3 and Table 4).

Fig. 3
figure 3

Comparison of landscape value proportions between the Akashi Bridge and Kurobe Dam projects. *** p value < 0.0001; * p value < 0.001

Full size image
Table 4 Comparison of the average proportion of landscape values (%) between the Akashi Bridge and Kurobe Dam projects
Full size table

For the cultural groups, the results showed a good level of consistency across broad aspects of each category, especially between the Japanese and English language groups, although for three of the categories, there were obvious differences of varying levels of significance between the Japanese and Chinese groups (see Fig. 4 and Table 5).

Fig. 4
figure 4

Comparison of landscape value proportions at Akashi Bridge among different tourist groups. *** p value < 0.0001; * p value < 0.001

Full size image
Table 5 Comparison of the average proportion of landscape values (%) of Akashi Bridge between different tourist groups
Full size table

Regarding the groups for the three bridges, there was also a significant difference among most of the categories relating to landscape values, but the variation was largely determined by the distinction between two of destinations by their infrastructure. For instance, the difference between the Akashi and Irabu bridges primarily refers to matters of form and practice or activity. This is only partially related to relationships; however, the difference between the Akashi and Rainbow bridges is entirely related to relationships, including aesthetics, landscape, social-cultural aspects, and tourism. The categories for the landscape values between the Irabu and Rainbow bridges are generally different, except for the image relationships and overall composition (see Fig. 5 and Table 6).

Fig. 5
figure 5

Comparison of landscape value proportions for each bridge structure. *** p value < 0.0001; * p value < 0.001

Full size image
Table 6 Comparison of the average proportion of landscape values (%) for each bridge structure
Full size table

5 Discussion

5.1 Landscape value distribution across different kinds of construction

Despite similar overall proportions, the composition of the landscape forms, including the juxtaposition of nature and structure, elicited nearly opposite values in the two projects (as shown in Fig. 6). The proportion of domestic visitors referring to natural aspects at the Kurobe Dam was much larger than that at the Akashi Bridge. Although the latter is not entirely lacking in natural elements, references to sea, sky, and sea wind seem to be perceived as somewhat monotonous, thus limiting the degree of natural appreciation. On the other hand, elements relating to water (lakes), mountains (peaks, red leaves), and weather (snow, rain) at the dam site offer pronounced environmental variety, significantly enhancing visitors’ appreciation of the natural environment.

Fig. 6
figure 6

Covariance difference between the Akashi Bridge and Kurobe Dam

Full size image

The conditions are reversed in regard to the actual structures. Whereas the value attached to nature is perceived more frequently in the dam project, the bridge structure is marked by a high degree of focus on the structure body itself, which means that the bridge, with its long span and high tower, impressed visitors much more than the dam did dam tourists. Previous studies have pointed out the function of bridges in representing places and its impact on the image of the local area. The bridge tended to be viewed as a self-contained symbol linked to the creation of a more meaningful place (Warnaby & Medway, 2008). Compared to the dam project, the bridge structure predominates in references to large-scale infrastructure is also presented from the visitors’ viewpoints as one of the landscape values that added to their experience.

The differences in the type of practice or activity were not as significant as they were for the structure or relationship subcategories. This suggests that considerable attention was given to activities and behaviors at both the dam and bridge projects. The tourists were interested in traversing the bridge deck from one end to the other and walking along the road, viewing the scenery from the deck. This validates the attraction targets set by the tour companies and was well accepted by the tourists themselves. Another possibility was sailing a boat and taking photos from a viewpoint at sea level. For visitors, the most interesting feature at the dam site was the water discharge process, an exclusive feature of the structural form. Walking up and along a mountain was also a popular activity. The proportion of practice-based values was somewhat higher for the dam project than for the bridge. As there were optional lake activities, the practices at the dam could match those at the bridge to an extent.

Aesthetics in relationships refers to the direct appreciation and enjoyment of the landscape and scenic views. It turned out that people expressed more landscape-related aesthetic values for the long-span bridge than they did for the dam, although the landscape is often appreciated as aesthetically valuable (Fox et al., 2016; Marianne et al., 2017), and the composition of water bodies, land cover, and mountains is considered to have high aesthetic value (Brown & Brabyn, 2012; Vouligny et al., 2009). In our study, the reference to landscape aesthetic values together with the structural landscape properties was more prominent in the bridge cases that were less marked by elements of nature. This mainly arises from the fact that the structures themselves have contributed much to landscape appreciation in the context of large-scale infrastructure tourism. This is also partly because of the night skyline and the panorama that can be observed from the top of the high tower. For place-based value, the bridge, by functioning as a link between two locations, prompted greater awareness of place for the visitors, while the dam project seemed less connected to a specific location.

From a sociocultural point of view, the Kurobe Dam generally offered more comprehensive information about its relation to society (as shown in Fig. 6), which was further strengthened by the mass media coverage of its construction. The tourism function of both structures appears to be given a similar degree of attention by the visitors (although the p value is quite small), being just slightly higher for the dam. Finally, relational feelings and images were bound up with a complex body of elements, with a variety of landscape and touristic possibilities leading to a more significant set of image values for the Kurobe Dam.

5.2 Landscape value distribution across different cultural groups

The effect of human cultural background on general landscape perception and the tourism experience has been previously discussed, and the results obtained have shown some inconsistencies (Bernatek-Jakiel & Jakiel, 2013; Jacobs, 2011; Yu, 1995; Zhang et al., 2015). Some scholars argue that cultural factors are the most influential determinant in landscape perception, while others report similarity of landscape perception and preference (Byoung-Eyang and Kaplan 1990). In our study, the results show that the differences among cultural groups are not based solely on the aesthetics of the landscape. The landscape experience at the large-scale bridge sites varied, particularly in relation to specific landscape values: structural form, practices, and sociocultural relationships.

The infra-tourism project at the Akashi Bridge site has successfully attracted both domestic and foreign visitor groups from different countries. Previously, tourism research has proposed theories to utilize cultural backgrounds as a factor of analysis (Hall, 1977; Hull & Reveli, 1989). Using this theoretical framework, studies grouped tourists from mainland China, Taiwan, Japan, South Korea, or Southeast Asia into an Asian group, while tourists from North America, Europe, Australia, or New Zealand were categorized into a Western group (T. Y. Choi & Chu, 2000). Likewise, we grouped online visitors and tourists per language use into an English-language group, coming mostly from English-speaking areas, and an Asian group including Japanese domestic and Chinese foreign visitors to explore the difference between these two groups in large-scale infrastructure tourism (as shown in Fig. 7).

Fig. 7
figure 7

Covariance difference among cultural groups

Full size image

Compared to their Japanese domestic counterparts, Chinese visitors displayed a significantly higher focus on the bridge as a structural form and on the sociocultural values attached to the structure (as shown in Fig. 7a). In contrast, the practice-based values perceived by the Japanese visitors suggested that they had more interest in site activities, including walking, boating, and photography, than the Chinese group. This may also relate to the extent of domestic tourists’ familiarity with the sites. It was reported in the results of a Japanese nature-based tourism survey of inbound foreign tourists that they could not accurately clarify what they experienced in a place (Jones & Ohsawa, 2016), which is attributed to the discrepancy in familiarity between domestic and foreign tourist groups (Funck & Cooper, 2013).

Similarly, a comparative analysis of Japanese domestic tourists has indicated that compared to foreign tourists, domestic tourists may be interested in a wider variety of aspects of tourism sites, leading to a wider range of locations to visit (Maeda et al., 2018). Infrastructural tourist destinations are sites that emphasize activity-based experiences. Although official guided tours offer the possibility of some activities, foreign visitors who are unfamiliar with local arrangements were quite restricted in their activity choices at the sites, while the Japanese group felt freer to enjoy the experience with the information obtained from the tour company’s promotional materials.

The most significant difference visible among the three cultural groups occurred at the level of sociocultural relationships. The Japanese domestic group paid the least attention to this aspect, whereas the Chinese visitors displayed the strongest interest in the sociocultural values associated with the bridge. They tended to refer more frequently to the effect and roles of the government and the country as a whole within the construction and management process. They were also more likely to express interest in how the building and construction work had been planned and conducted, especially with regard to earthquake protection. This may derive from images of large-scale construction work recently undertaken in China, and thus, their interest in infrastructure is also an expression of a stronger awareness of their political and social lives in relation to power and collective consciousness (Luo et al., 2013). A recent analysis of Chinese tourism sentiment at an Australian destination also reported higher interest in the sociocultural aspects of sites, expressed by the frequency of occurrence of the terms “famous” and “ancient” (Liu et al., 2019). In another study, Chinese visitors tended to express their opinions of a domestic large-scale hydraulic project most frequently in terms of political meaning and tourism functions at the same time (Jiang et al., 2016).

The differences in landscape experience between the Asian- and English-language groups become more unified regarding most landscape values, reflecting a relatively similar appreciation pattern across cultures in the case of the large-scale bridge tourism site. However, different tendencies can still be seen, particularly in terms of the form of structure and tourism elements, which were more appreciated by the Asian- and English-language groups, respectively (as shown in Fig. 7b). The Asian group was generally more willing to appreciate the bridge presented as an integral structure, or sum of components, than their English counterparts. Similar results to the English group were also found among the Chinese tourist group, who paid more attention to specific landmarks rather than natural attractions at general destinations (Liu et al., 2019). However, the English group was more interested in tourism practices and activities, including restaurants and hotels, and the journey to the destination. Some studies have identified English culture as a low-context culture, valuing verbal skills and communication through talking, which are related to tourism elements (Liang, 2010). For them, the large-scale bridge and the surrounding environment are more likely to be of equal attraction rather than favoring the values provided by the structure itself.

5.3 Landscape value distribution across different locations and environments

As with the results for the two distinct structural forms, the perceived landscape values also showed notable differences according to location and environment for bridges of the same structural form (see Fig. 8). The specific variations presented quite distinct features across different pairs. Considering that the Akashi area is one of the common local regions inside Japan, it is possible through comparison of the other two bridge sites to the Akashi Bridge to see the effect of the urban center environment and the fame of the tourist destination on the landscape experience at the large-scale bridge sites.

Fig. 8
figure 8

Covariance difference among bridges located in different environments. a Between Akashi Bridge and Irabu Bridge. b Between Akashi Bridge and Rainbow Bridge

Full size image

There were higher levels of appreciation for natural forms in relation to the Irabu Bridge in Okinawa, where ordinary natural elements such as the sea, weather, or sky and distinctive natural elements such as typhoons, sea turtles, and coral reefs attracted more attention than Akashi Bridge (as shown in Fig. 8a) or Rainbow Bridge. Correspondingly, the importance of value pertaining to the bridge structure itself decreased at tourist sites such as Okinawa. This reveals that characteristic expectations relating to tourist destinations have an impact on visitors’ perceptions of infrastructural sites. An interesting tendency can be found at both Rainbow Bridge and Irabu Bridge regarding landscape value and practices. In addition to everyday activities, such as walking and boating, the most distinctive practices performed by domestic visitors at tourist destinations were connected to driving experiences. A large proportion of the activities associated with the Irabu Bridge revolve around automobiles: driving across the bridge, parking the car, finding a parking space, and navigating one’s way there in the first place (as shown in Fig. 8a). Some people also mentioned cycling. At the Rainbow Bridge, however, walking and cycling are normal activities undertaken by everyone as a normal part of city life.

It is the surrounding urban environment rather than the tourism destination itself that contributes much more to the comprehensive landscape aesthetics in the case of the large-scale Rainbow Bridge. Located in the urban center, Rainbow Bridge was perceived to have higher aesthetic value by visitors mainly due to the landscape and scenic beauty (as shown in Fig. 8b). When combined with values associated more broadly with landscape properties, it is obvious that, compared with the other sites, the Rainbow Bridge is famous not for its scale and size but rather for the scenic views of the surrounding urban landscape.

Relationships regarding the sites also displayed some variation. Comparatively, the Akashi Bridge was perceived to have the strongest reference to place because of its function and position. The tourist destination of Okinawa and the urban settings in Tokyo, with a more explicit identification, by contrast, direct any specific visitor to focus on places and locations. Once again, it was confirmed that Japanese domestic visitors are not as concerned with the sociocultural aspects of the constructions as their counterparts from overseas, especially those who come from neighboring areas. Regarding the same structural forms, the effect of organized tours at the Akashi Bridge is noticeable because it promotes a wider set of tourist provisions such as accommodations and transport. Finally, relationships associated with images and feelings proved to be almost the same across the three bridges, suggesting that the same structural form has a similar impact on emotional reflection and spiritual expression.

5.4 Implications for landscape and tourism practices in terms of large-scale civic infrastructures

To ensure the protection of diversity and originality, landscape assessments of infrastructure proposals have been discussed and investigated (Chen et al., 2018; Marianne et al., 2017), while discussions about the effect of existing infrastructure projects on people and areas await deeper investigation. This study provides useful insights into the relationship between landscape value and the perception of large-scale civil infrastructure among visitors. The landscape value of a given site can show significant differences under the influence of factors such as the type of construction, location and environment, and different cultural backgrounds, while different factors can affect the specific landscape value to different degrees (as shown in Fig. 9). For civil infrastructure, structural form is probably still the most important aspect, as it almost determines the way the public perceives and experiences it. The influence of location is strongly related to the specific context in which the destination object is situated, while cultural groups, with different sociocultural focuses, differ even more in terms of activities and tourist behavior. The diverse opinions studied here can help us identify values often overlooked by general consideration. A better understanding of the heterogeneity of landscape experiences at tourism sites is also important for their management (K.-C. Lee & Son, 2017).

Fig. 9
figure 9

The degree of effect on landscape values among different factors. Shades of color mapped by covariance differences express the degree of intergroup differences caused by different factors in each category (deeper color with more significant variance)

Full size image

From a practical perspective, large-scale civil infrastructure is a comprehensive subject with a variety of physical, social, political, and cultural attributes, and the presence on large scale can have a significant impact on the form of regional space, as well as on key socioeconomic effects in terms of its function. The tourism function is a potential approach to sustainability, especially for constructions that have an intellectual property (IP) effect or where the original function is in decline. From this perspective, an understanding of landscape values, as well as the associated ecosystem services and social utility study, is of great interest.

As the Japanese Minister of Tourism has promoted a plan for Japan to become a “tourism-focused country,” the Japanese government and authorities have focused on tourism development, especially the inbound tourist market. In this study, significant variation was found in the comparative analysis of perceived landscape values bound up with large-scale infrastructure, which are in turn closely linked to landscape and tourism management practices. A considerable proportion of the various practices and tourist values expressed clarify the effectiveness of local authorities in providing official guided tours at the Akashi Bridge. Taking the coincidence of symbolic structural form with the high degree of place reference is a good starting point for local authorities to promote the regional image as more deeply connected with the structure itself, as in other cases (Warnaby & Medway, 2008). At the same time, the relatively low interest shown in practice-related values by the overseas group indicates that it would be beneficial to improve internationalized services to offer better guidance to visitors unfamiliar with local practices. Many English-language visitors referred to the language barrier at tourism sites. A more understandable and user-friendly introduction to the background and history of the various structures would also promote higher levels of sociocultural interest.

Dam visitors would also benefit from collaboration with environmental managers and local authorities to improve the overall experience for those interested in either the natural or the structural aspects of the environment. It would be beneficial to integrate tourism facilities covering the surrounding mountain and watershed area into the dam tourism site, facilitating a wider range of license development in local communities. The impact of practice aspects such as the parking issues at Irabu Bridge also underscores the task of pushing tourism destination administrators to provide more adequate facilities to support tourism around the structures. In the large-scale infrastructure tourism space, developed structures appeal to visitors and tourists, and economic growth related to promoted facilities in the area and their development in turn adds to the uses of structural tourism and related investments (Włodarczyk, 2009). Identifying variations related to construction and location would help companies and authorities reconsider how to utilize structures and the corresponding environmental features. This may be particularly useful in considering the relationship between structure and location to promote more sustainable and persistent tourism attractions for both domestic and inbound markets, which has been extensively discussed in the Japanese literature on regional revitalization and sustainable development through infra-tourism and similar practices.

The online UGC content in this study was used to evaluate landscape values provided by large-scale infrastructure sites and to understand the perception and preferences of infra-tourists (Richards & Friess, 2015). The data and approach used here can potentially be used to assess the intersection between structural facilities and the surrounding environment and the potential to benefit infra-tourism development, which is a growing focus of both researchers and policy-makers in sustainability tourism (Donahue et al., 2018). The study has also revealed certain advantages pertaining to methodologies that draw on UGC data: (1) There is abundant data regarding the perceptions and assessments of various groups that can be obtained at little or no cost; (2) UGC provides valid, informative, and authentic perceptions and feelings that arise from actual experiences; (3) the analysis process has no influence or effect upon the study subjects; and (4) calculations based on the TF-IDF weightings for words can generate a quantitative representation of content and its distribution related to theoretical categories of landscape values that in turn make possible a statistical analysis of the differences between various groups.

A combined qualitative and quantitative content analysis approach was developed, which consisted of two phases. In the first phase, computerized text processing was implemented to prepare the data for qualitative analysis, and a theory-oriented categorization process was used to obtain the proportion of landscape values in each category as the unit of analysis. In the second phase, for the unit of analysis, statistical methods including intergroup tests and covariance analysis were used to quantify differences in landscape values under different external factors. The approach to text in this paper stems from an adaptation of structural topic modeling (Roberts et al., 2014; Roberts, Stewart, and Tingley 2019), but we use theoretical constructs as the source of the topics rather than automatically generated categories based on probability. This provides a practical way of combining traditional content analysis with computerized text processing approaches.

The content analysis and discussion were generally based on a CVM theoretical framework, along with some additions and modifications derived from other related studies. The consideration of landscape values offers the possibility of close links with other theories, such as the cultural ecosystem services theory mentioned briefly at the outset (Smith & Ram, 2017). Future studies could, therefore, investigate the applicability and possible interrelationship between these two approaches regarding large-scale infrastructure and other objects involving tourism functions and landscape features.

5.5 Limitations and perspectives

Some study limitations should be noted, mainly resulting from time constraints and the research approach. First, in analyzing the impact of different visitor groups on the landscape values at the Akashi site, we acknowledged the choice of two platforms for enriching the amount of data from domestic reviews in China. There are differences in the representativeness of the data across these platforms. Since online reviews from different platforms may differ or be potentially biased in terms of linguistic characteristics, semantic features, affective tendencies, and impact on users (**ang et al., 2017), the variance of reviews on different platforms should be further considered to obtain a more accurate result. Regarding the study methodology, the categorization system had to be developed by two individuals to avoid any bias arising from personal judgments. Previous research has reported that statistical tests are more reliable for building categorization systems if more people are involved in the development process (Oteros-Rozas et al., 2018).

Alternative processing approaches to the segmentation of phrases at the multiword level can provide a better interpretation of contextual meanings and word combinations than single words can. Additionally, calculations to obtain the representativeness of the review contents were realized according to a single weighted value for the TF-IDF of each word within the categories. A vectorized representation of the words would work as an alternative technique to enable a more precise expression of content. Machine learning methods such as neural networks can be used to handle various types of text generation or categorization tasks very well. Especially when the number of samples is large enough, computers can automate the processing of large amounts of data for broader and larger-scale analysis. Computers also have perfect reliability, and thus, intercoder reliability is not an issue. Nonetheless, the data used in this study are more like a survey sample collected unconventionally, and the limited amount of data may lead to doubts about the validity of using machine learning involving semantic understanding. Especially for short review text with high randomness and variability, it is more difficult for language processing algorithms to analyze the content (Rashid et al., 2019) than they can for the more standardized and formal text corpus (e.g., newspapers, essays). In addition, the results obtained by machine learning, especially large-scale neural networks, still cannot be fully understood or interpreted by humans (Arrieta et al. 2020).

In terms of the research question, this study is more similar to computer-assisted qualitative content analysis combining quantitative methods than fully quantitative text mining. The topics generated by topic modeling approaches were found to be hard to interpret or to express clearly. Considering that landscape value, as a socially constructed concept, is inherently difficult to characterize, the use of a theory-driven approach is necessary and meaningful. For example, for the massive scale of short tweets, many unsupervised learning methods are used for the analysis and prediction of specific information in the text (e.g., based on location information), with less focus on the relationship between specific textual semantics and the measurement of specific social constructs.

Moreover, current large language models (LLMs) based on deep learning exhibit textual representation and semantic comprehension capabilities beyond what has been achieved in the past. In the future, LLMs may be applied in different ways to study social topics (e.g., to measure political ideologies (Wu et al., 2023)) or to change traditional scientific research methods (e.g., text-based causal inferences (Egami et al., 2022)).

6 Conclusions

The landscape cultural value model related to cultural ecosystem services used here provides an effective tool for the potential assessment of tourism functions and tourism-focused landscape features for infrastructure sites, which have recently emerged as prime visitor destinations in Japan. In this study, we gathered online reviews regarding large-scale infrastructure from visitor groups with different cultural backgrounds. A quantitatively grounded representation of review content provided the basis for a comparative statistical analysis. The results revealed the following: (1) The landscape values relating to the infrastructures are influenced by the actual structural forms, with environmental variety being more significant for the dam and form regularity for the bridge projects. (2) Different visitor groups perceived landscape values with different preference levels according to their background. Having similar landscape aesthetics, domestic groups in Japan experienced more through tourist activities but were somewhat indifferent to sociocultural aspects, while for other tourist groups, this was reversed. (3) The characteristics of locations and environments were also seen as having a significant influence on perceived landscape values. Thus, the configuration of an urban setting can promote higher levels of aesthetic value, while the identity of a tourist destination can undermine a specific sense of location and place. This paper explored the landscape values associated with massive infrastructural projects in the tourism context and demonstrated that it is methodologically feasible to perform a quantitative analysis grounded in this kind of framework. These results are worth considering in future studies for a comprehensive understanding of tourism and landscape in related fields, encouraging relevant authorities to reconsider sustainable tourism practices and spark a regional renaissance across a range of possible projects.