2.1.1. Research on destination image.

Destination image (DI) is generally understood as the comprehensive mental representation that visitors form of a destination’s resources, environment, services, and experiences. Since Hunt introduced the destination image concept into tourism studies [10], research in this field has gradually developed an analytical approach centered on cognitive image, affective image, and behavioral intention [11]. Echtner and Ritchie extended the measurement of destination image through dimensions such as attribute–holistic, functional–psychological, and common–unique components [12], while Baloglu and McCleary proposed the “cognitive–affective–overall image” model [13]. These studies indicate that destination image research is well-equipped with multidimensional frameworks; existing theories can effectively explain visitors’ attribute cognition, emotional evaluation, and behavioral tendencies toward general destinations.

However, classical DI models have mainly been developed for ordinary natural scenic areas or macroscale urban destinations [14,15]. Their systemic frameworks are largely built upon visitors’ psychological processes and evaluations of destination attributes, frequently overlooking the specific spatial structures, social activities, and local cultural contexts that anchor these perceptions. For living HCNs, the object of visitor perception is not merely an isolated attraction, commercial service, or cultural symbol. Rather, it is a composite field shaped by heritage spaces, community life, recreational consumption, local memory, and symbolic meanings. If the “cognitive–affective–conative” framework is rigidly applied, the tourism image of HCNs risks being simplified into mere attribute evaluation, emotional tendency, or behavioral outcome. Consequently, this framework fails to adequately explain how visitors’ perceptions are triggered and organized within specific spatial hierarchies and living contexts, and how they gradually evolve into emotional identification and meaningful resonance. Research on the tourism image of living HCNs, therefore needs to build on existing DI theory while incorporating the analytical perspectives of spatial hierarchy, situated experience, and meaning generation.

2.1.2. UGC-based research on tourism image perception and its limitations.

With the development of online travel platforms and social media, user-generated content (UGC) has become an important data source for research on the perception of tourism image [16]. Compared with questionnaires and interviews, UGC offers several advantages, including more natural expression, stronger contextuality, and broader coverage. It can capture visitors’ spontaneous attention and emotional responses during actual travel experiences and subsequent online sharing. In recent years, related studies have moved beyond word-frequency statistics and content analysis to include quantitative approaches such as text mining, semantic network analysis, sentiment analysis, and IPA [17–20].

However, the use of UGC does not in itself ensure greater explanatory depth. Some studies remain confined to a largely descriptive level of “high-frequency word extraction–theme classification–sentiment judgment,” treating transportation, architecture, commerce, folk customs, and emotions as parallel indicators. This approach often fails to adequately explain the spatial embeddedness, hierarchical relationships, and planning implications of different perceptual elements. Without the guidance of a systematic hierarchical theory, UGC analysis can easily be reduced to a mere enumeration of high-frequency words and fail to reveal the structural interactions among physical spaces, social settings, cultural contexts, and conceptual meanings. Therefore, the use of UGC in this study does not seek to highlight the novelty of text mining itself. Rather, it aims to integrate fragmented visitor perceptions from UGC into a codable and comparable hierarchical analytical framework, thereby providing a basis for the subsequent construction of the evaluation system and multi-case comparison.

2.1.3. Specificity of tourism perception in living HCNs.

Unlike highly scenic area-oriented tourism spaces, living HCNs retain the everyday routines of long-term residents, local community relations, and living cultural heritage. Their tourism image is not determined solely by visual symbols or commercial formats, but is dynamically generated through multidimensional interactions among “human–space–activity–meaning” [21]. Existing research on HCNs has generally followed three lines of inquiry. The first focuses on conserving street fabric, revitalizing public spaces, and organizing visitor routes, emphasizing the preservation, renewal, and functional efficiency of historic spaces [22–25]. The second examines commercial restructuring, heritage interpretation, and authenticity-based experiences, with attention to how tourism development affects local cultural representation and everyday neighborhood life [26,27]. The third explains tourism experience outcomes from the perspectives of tourist satisfaction, place attachment, and behavioral intention [8,28]. These studies provide an important basis for understanding the conservation and use of HCNs. However, the current literature still lacks an explanatory framework that systematically integrates spatial hierarchy, living context, visitor perception, and the mechanisms underlying the formation of tourism image in HCNs. Comparative analysis across different types of living HCNs also remains largely underexplored.

2.2.1. Structural differences between the HSSPR framework and classical models.

Compared with classical DI models, the HSSPR framework differs primarily in three aspects: explanatory perspective, hierarchical logic, and data organization. First, classical models tend to emphasize changes in tourists’ psychological processes, whereas the HSSPR framework re-embeds psychological perception within the physical and social spatial hierarchy of HCNs, highlighting the spatial embeddedness and contextual activation of tourism image formation. Second, classical models often focus on the linear influence of attribute evaluation on behavioral intention, whereas HSSPR is designed specifically for living HCNs and can capture and explain the progressive transformation along the sequence of “material space–setting participation–cultural empathy–spiritual resonance”. Third, in response to the highly fragmented nature of perceptual content in UGC texts, the HSSPR framework uses the four layers of Space, Setting, Place, and Realm to provide a classification and coding system with a rigorous hierarchical logic, as well as a quantitative diagnostic pathway for large-scale unstructured online texts. This helps avoid the simplistic juxtaposition of heterogeneous perceptual content.

In summary, HSSPR can be operationalized as a codable, comparable, and preliminarily testable analytical tool. This study uses it to construct an analytical and evaluation framework for tourism image perception in living HCNs. By comparing the word-frequency, semantic, and sentiment characteristics of UGC across the three cases, the study identifies shared perceptual features and type-specific differences across neighborhoods, and uses IPA quadrant analysis to diagnose the strengths and key constraints that limit tourism image enhancement.

3.2.1. Network text analysis and deep-learning-based sentiment computation workflow.

Network text analysis can extract visitors’ perceptual themes, sentiment tendencies, and evaluative elements from user-generated content (UGC), making it suitable for identifying the perceptual structure of tourism image in living HCNs. Compared with traditional questionnaires, UGC provides a larger sample size and more closely reflects visitors’ spontaneous expressions. It can therefore supplement the analysis by capturing immediate feelings, spatial evaluations, and emotional feedback during tourism experiences. Based on the hierarchical HSSPR framework, this study developed an analytical workflow consisting of “data collection–text cleaning–sentiment recognition–element matching–IPA diagnosis”. First, multi-platform online data were collected for the three cases: Longwei Pass, Santai Mountain Historic Area, and Ming-Qing Tonghai Old Town. Second, the raw texts were processed through deduplication, word segmentation, stop-word filtering, removal of irrelevant information, and expression standardization, generating a valid analytical corpus. Third, by combining sentiment recognition results with a keyword feature lexicon, visitor comments were mapped onto specific perceptual elements under the HSSPR framework. Finally, the frequency of element mentions was used to represent attention salience, while sentiment evaluation results were used to represent performance. These indicators were then incorporated into the adaptive IPA model to identify perceptual advantages, experiential deficiencies, and optimization priorities across the different neighborhoods.

In the sentiment recognition stage, given that short online reviews are inherently characterized by colloquial language, fragmented expression, and strong context dependence, relying solely on sentiment dictionaries or traditional machine-learning methods may be inadequate due to insufficient contextual adaptability and a limited ability to identify implicit emotions. Therefore, this study built a text convolutional neural network (TextCNN) model for sentiment tendency recognition using Python and the PyTorch framework. To improve the model’s adaptability to the context of tourism reviews on HCNs, the training corpus was drawn from historic towns or neighborhoods in Yunnan that share similarities with the three empirical cases in expressions related to historical remains, traditional streets and alleys, local life, and tourism experience. These sources included Heijing Ancient Town, Weishan Ancient City, Shaxi Ancient Town, and Lijiang Ancient City. The training corpus covered UGC from online platform reviews, Xiaohongshu posts, and related comments, encompassing common expression scenarios such as historic sites, street-and-alley spaces, local dining, commercial consumption, transportation services, physical and mental experiences, and recommendation intentions.

On this basis, model training followed a weakly supervised learning procedure of “rule-based annotation–manual review–TextCNN”. First, review texts were preliminarily labeled as positive, neutral, or negative according to rules based on sentiment dictionaries, degree adverbs, negation words, and contrastive structures. The preliminary labels were then manually reviewed, with particular attention to correcting misclassified samples involving irony, mixed emotions, very short sentences, and ambiguous sentiment expressions. This process produced a deduplicated labeled dataset of 7,532 records, including 4,035 positive, 2,766 neutral, and 731 negative samples. To test the reliability and robustness of the model recognition results, the labeled dataset was divided through stratified sampling into a training set of 6,025 records, a validation set of 753 records, and a test set of 754 records, while maintaining the same proportional distribution of the three sentiment labels. The model used character-level encoded texts as input. The maximum text length was set to 160, and the character vocabulary size was 3,618. The main parameters included an embedding dimension of 64, 64 convolution kernels, convolution kernel sizes of [2–5], a batch size of 64, and 10 training epochs. The TextCNN model extracted local semantic combination features from short texts and used a Softmax layer to produce the probability distribution over sentiment categories. The test-set results showed a classification accuracy of 86.34%, a macro-precision of 80.26%, a macro-recall of 78.28%, and a macro-averaged F1 score of 0.7918, indicating that the model performed well in identifying sentiment in short tourism reviews related to HCNs. The trained model was then applied to batch sentiment recognition of comments from the three study sites, providing a basis for subsequent performance measurement and IPA analysis.

3.2.2. Importance–performance analysis.

To incorporate perceptual attention and sentiment evaluation from online texts into a unified diagnostic framework, this study adapts the traditional importance–performance analysis (IPA). The frequency with which an element is mentioned in online reviews is not fully equivalent to subjective importance measured through questionnaires. Therefore, the IPA results in this study are used primarily to identify mismatches between “attention” and “experience” in tourists’ online expressions, rather than to establish strict causal relationships. In this study, “importance” is treated as a proxy indicator of attention salience in tourists’ perceptions. “Performance”—conceptualized here as experiential performance—is calculated from the sentiment probability distribution generated by the TextCNN model, serving to represent tourists’ experiential feedback on the relevant perceptual elements.

First, the importance score of each perceptual element was calculated. For a given case, if the -th review was matched to the -th perceptual element, then =1; otherwise, =0. The occurrence frequency of the -th perceptual element was calculated as follows:

(1)

Where denotes the occurrence frequency of the -th perceptual element in a given case, and denotes the number of valid experiential evaluation texts in that case. The proportion of the occurrence frequency of this element to the total occurrence frequency of all perceptual elements was then used as its importance score:

(2)

Where denotes the importance score of the -th perceptual element, and denotes the total number of core perceptual elements extracted through text mining and grounded coding. The denominator represents the total occurrence frequency of all perceptual elements in the given case.

Second, the performance score of each perceptual element was calculated. To avoid obscuring intensity differences within the same sentiment category through the discrete assignment of “positive = 5, neutral = 3, negative = 1”, this study constructed a continuous sentiment performance score for each review based on the three-class sentiment probabilities output by the TextCNN model:

(3)

where denotes the sentiment performance score of the -th review; and denote the probabilities of positive, neutral, and negative sentiment output by the TextCNN model, respectively, with their sum equal to 1. The performance score of the -th perceptual element was represented by the mean sentiment performance score of all reviews matched to that element:

(4)

Finally, IPA quadrants were constructed using a dynamic mean-splitting method. The mean importance score and mean performance score of all perceptual elements in a given case were calculated and used as the dividing axes of the four-quadrant matrix:

(5)

(6)

According to the position of each element’s coordinates () relative to the mean lines, the elements were classified into four quadrants: maintain advantages, priority improvement, moderate maintenance, and low priority:

(7)

Where denotes the quadrant assignment of the -th perceptual element. Given that online UGC texts generally show a concentration of positive sentiment and relatively high overall performance scores, this study used dynamic dual means rather than fixed midpoint values as the dividing lines. This approach allows the matching or mismatch between attention salience and experiential performance to be identified more effectively.

3.3.1. Multi-platform online data collection.

This study selected Longwei Pass, Santai Mountain Historic Area, and Ming-Qing Tonghai Old Town as empirical study sites. As the target users were mainly domestic Chinese tourists, online textual data were obtained from three major Chinese online platforms: Dianping (https://www.dianping.com), Ctrip (https://www.ctrip.com), and Xiaohongshu (https://www.xiaohongshu.com). These platforms are respectively characterized by consumer reviews, travel evaluations, and social media sharing, and thus can reflect visitors’ spatial experiences, consumption perceptions, and overall impressions of living HCNs from different perspectives.

Given the differences in data structure and textual presentation across platforms, this study adopted platform-specific collection strategies. Data from Ctrip were obtained using Python Requests to access publicly available web interfaces, through which comment text, posting time, user ratings, and related fields were collected. Data from Xiaohongshu were collected using Selenium-based automated web scraping, targeting post titles, body text, and multi-level comment texts. Data from Dianping were obtained using Reqable, a network protocol analysis tool, to parse publicly available review data from mobile H5 pages and WeChat mini-programs, extracting fields such as review text, posting time, and star ratings. To ensure proper and ethical use of data, only publicly available texts were collected, and no private communications or sensitive information were involved. User identifiers were used only for deduplication and sample validation and were excluded from subsequent statistical analyses.

After multi-platform collection and preliminary aggregation, a total of 4,219 initial online textual records were obtained. These included 1,619 records for Longwei Pass, 914 records for Santai Mountain Historic Area, and 1,686 records for Ming-Qing Tonghai Old Town. The volume of UGC was generally consistent with the degree of tourism development across the cases. The distribution of raw texts across platforms and cases is shown in Table 1.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Number of raw online textual records collected by the platform.

https://doi.org/10.1371/journal.pone.0355218.t001

3.3.2. Text preprocessing and sample screening.

The raw corpus collected from multiple platforms contained inconsistencies in formatting, duplicate content, and a mixture of texts unrelated to tourism experiences. To ensure the reliability of subsequent sentiment recognition, perceptual-element coding, and IPA analysis, the raw texts underwent standardized preprocessing and sample screening.

First, analytical units were reconstructed, and formats were cleaned across platforms. For platforms such as Xiaohongshu, where posts may include titles, body text, and multi-level comments, relevant content was reorganized into independent textual units. Fields such as the original platform source, comment level, original row number, and comment text were retained to ensure data traceability. Python scripts incorporating regular expressions were then used to clean the texts by removing noise such as whitespace characters, web links, platform tags, garbled characters, and emojis without clear semantic meaning. Exact duplicates were also removed. After preprocessing, 3,800 textual records were obtained across the three cases, including 1,506 for Longwei Pass, 843 for Santai Mountain Historic Area, and 1,451 for Ming-Qing Tonghai Old Town.

Second, texts not associated with tourism experiences were classified and screened. Comment sections on platforms such as Xiaohongshu often contain information-seeking texts about locations, routes, or photo spots; social reply texts such as emojis, greetings, and expressions of thanks; as well as short texts with limited information and texts expressing only interest arousal. Although these texts may reflect online engagement, they usually lack substantive evaluations of historical landscapes, spatial environments, facilities and services, local life, or recreational experiences. If included directly in the analysis, they could artificially inflate the proportion of neutral texts and affect the interpretation of IPA results. Therefore, this study combined keyword matching with manual review to exclude texts unrelated to tourism experiences. Only texts that explicitly addressed perceptual content related to landscapes, services, transportation, culture, spatial atmosphere, consumption experience, or local life were retained as valid samples. A total of 2,540 valid experiential evaluation texts were finally included in the main sentiment analysis and perceptual-element matching, including 1,020 for Longwei Pass, 536 for Santai Mountain Historic Area, and 984 for Ming-Qing Tonghai Old Town. The detailed screening results are shown in Table 2.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Results of text preprocessing and valid sample screening.

https://doi.org/10.1371/journal.pone.0355218.t002

4.1.1. High-frequency word analysis: historical continuity, spatial environment, and everyday life jointly shape tourism image.

The top 20 high-frequency words indicate that visitors’ perceptions of the three sites primarily center on historical resources, the spatial environment, and everyday life experiences (Table 3). For Longwei Pass, high-frequency words such as “ancient city”, “history”, “historic site”, and “Nanzhao” point to memories of the historic pass and ancient city. Meanwhile, words such as “daily life”, “hustle and bustle”, “local people”, “coffee shop”, and “delicious” indicate that visitor perception has extended beyond viewing historical remains to include strolling through streets and alleys, everyday marketplace life, and light consumption experiences. The occurrence of “commercialization” also suggests that the involvement of tourism-related commercial formats may affect perceptions of authenticity. Therefore, the perceptual characteristics of Longwei Pass are mainly reflected in the image of a living neighborhood, continuously shaped by mountainous border-pass streets, and by alleys and everyday marketplace life.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Top 20 high-frequency words of the three sites.

https://doi.org/10.1371/journal.pone.0355218.t003

For Santai Mountain Historic Area, the dominant high-frequency words include “park”, “mosque”, “Confucian Temple”, “Bing Xin”, and “Lin Huiyin”. These terms indicate that its image is jointly supported by mountainous public space, religious culture, and memories of notable figures. However, the distribution of high-frequency words is relatively dispersed, suggesting that its core tourism symbols are not yet prominent. Visitor perception depends more on the interplay among scattered cultural nodes and public recreational spaces. At the same time, words such as “daily life” and “environment” indicate that everyday life in surrounding communities remains an important context for visitor perception. Compared with Longwei Pass, Santai Mountain Historic Area has not yet developed a mature tourism industry or continuous commercial visitor routes. Its tourism image is mainly shaped by the juxtaposition of Santai Mountain Park, the surrounding community life, religious, cultural, and educational nodes, and public recreational activities.

For Ming-Qing Tonghai Old Town, words such as “Xiushan Mountain”, “park”, “Confucian Temple”, “ancient city”, and “Qilu” form a relatively stable image of a mountain–water ancient town. Words such as “delicious”, “rice noodles”, “local cuisine”, and “daily life” further strengthen the perception of local life and everyday consumption experiences. Compared with Longwei Pass and Santai Mountain Historic Area, the keywords for Ming-Qing Tonghai Old Town are more concentrated around its mountain–water ancient-town structure, Confucian Temple culture, and local life, indicating that its tourism image rests on a relatively strong and integrated resource base. Overall, visitor perceptions across the three sites reveal the interweaving of history and culture, spatial environment, and everyday life, while also showing differences in the organization of everyday life and perceptual structure.

4.1.2. Semantic network analysis: hierarchical associations and case differences across space, setting, and meaning.

Based on high-frequency word statistics, word co-occurrence networks were further constructed from visitor comments across the three sites to identify the associative structure of perceptual content. The integrated semantic network shows that the tourism image of the three sites is not formed by a single resource or isolated keywords. Instead, it is organized through multilayered associations among history and culture, the spatial environment, everyday life experiences, and emotional evaluations (Fig 4(a)). Words such as “ancient city”, “history”, “Confucian Temple”, “architecture”, and “park” constitute visitors’ basic spatial recognition of the neighborhoods. Words such as “snacks”, “coffee”, “mountain climbing”, “photography”, and “distant view” point to specific processes of recreation and consumption. Words such as “treasure”, “delicious”, “openness”, and “commercialization” reflect visitors’ subjective evaluations of spatial atmosphere, experience quality, and development status. These findings suggest that the tourism image of living HCNs does not remain at the level of resource recognition, but forms associations among spatial objects, activity settings, and local meanings.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Semantic co-occurrence networks of high-frequency keywords.

https://doi.org/10.1371/journal.pone.0355218.g004

The independent semantic networks across the three cases further reveal distinct patterns of perceptual organization. In the network of Longwei Pass, historical and cultural terms such as “ancient city”, “historic site”, “Nanzhao”, and “history” are closely associated with experiential terms related to everyday life, such as “hustle and bustle”, “daily life”, “local people”, “coffee shop”, and “delicious”. This indicates a strong continuity among the historic pass remains, the street-and-alley scale, and everyday marketplace life, through which visitors can move relatively easily from historical recognition to daily experience (Fig 4(b)). The semantic network of Santai Mountain Historic Area mainly revolves around words such as “park”, “mosque”, “Confucian Temple”, “Bing Xin”, “Kuige Pavilion”, and “worship”. This suggests that its tourism image is mainly supported by mountainous open space, religious sites, cultural and educational nodes, and surrounding community life, although the semantic associations among these resources remain relatively dispersed (Fig 4(c)). In Ming-Qing Tonghai Old Town, words such as “Xiushan Mountain”, “Confucian Temple”, “ancient city”, “Qilu”, and “scenery” form composite associations, indicating that its tourism image is built through the combined effects of the mountain–water structure, ancient-town historical continuity, and local life. At the same time, terms related to natural landscapes, ancient-town culture, and everyday life show a multi-centered juxtaposition, suggesting that visitor perception is more strongly characterized by the holistic recognition of composite resources than by a concentrated orientation toward a single theme (Fig 4(d)).

Overall, the semantic network results can be interpreted hierarchically through the HSSPR framework. The differences among the three types of cases are not merely differences in resource types; rather, they reflect differences in spatial form, the embeddedness of community life, and visitors’ perceptual pathways. These results provide a semantic basis for the subsequent sentiment analysis and IPA diagnosis.

4.1.3. Sentiment analysis: sources of positive experience and deficiencies in service support.

Sentiment analysis was used to further identify visitors’ evaluative tendencies toward the tourism image of the three sites and potential deficiencies in it. Overall, comments on the three sites were dominated by positive sentiment: 1,866 comments were classified as positive, accounting for 73.46%; 438 were neutral, accounting for 17.24%; and 236 were negative, accounting for 9.29% (Table 4). Among the three sites, Longwei Pass had the highest proportion of positive comments, at 77.45%. Santai Mountain Historic Area recorded a positive sentiment rate of 67.72% and the highest proportion of neutral comments, suggesting that its valid experiential reviews contained more descriptive and weakly evaluative expressions. Ming-Qing Tonghai Old Town had a positive sentiment rate of 72.46%, but its proportion of negative comments was relatively high. Overall, the three sites all showed a solid foundation for visitor recognition, but differences in sentiment structure also indicate that the formation and consolidation of tourism image varied across cases.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Statistical results of sentiment analysis.

https://doi.org/10.1371/journal.pone.0355218.t004

The high-frequency words associated with positive sentiment show that positive evaluations of the three sites were all related to historical and cultural atmosphere, local living settings, and spatial experience (Table 5), although their sources differed. Positive sentiment toward Longwei Pass depended on the continuous interaction between historic pass remains and everyday marketplace life. For Santai Mountain Historic Area, positive sentiment was mainly supported by religious, cultural, and educational nodes, mountainous public space, and the atmosphere of community life. Ming-Qing Tonghai Old Town was associated with the mountain–water ancient-town structure, Confucian Temple culture, and local dining experiences. These results suggest that positive sentiment across the three sites was not triggered by a single historical resource, but instead relied on different perceptual structures characterized by continuity, complexity, and integrity.

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Statistics of high-frequency words associated with positive sentiment.

https://doi.org/10.1371/journal.pone.0355218.t005

Negative evaluations were more often directed toward problems related to transport accessibility, public service facilities, signage and guidance, visitor route organization, and certain consumption experiences (Table 6). This indicates that visitor dissatisfaction was concentrated mainly in intermediary links such as spatial access, service capacity, and cultural interpretation.

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. High-frequency words associated with negative sentiment and a summary of problems.

https://doi.org/10.1371/journal.pone.0355218.t006

4.3.1. Importance–performance scores.

Based on the IPA results, tourism image perception across the three sites showed shared strengths, common deficiencies, and type-specific differences among the HSSPR-coded perceptual elements (Table 9). In terms of shared strengths, C3 (historical and cultural relics) was located in Quadrant I (high attention–high performance) in all three cases, indicating that this perceptual element remains a core basis for positive evaluations of living HCNs. In terms of common deficiencies, B1 (food and snacks) was located in Quadrant IV (high attention–low performance) in all three cases. This suggests that although local food is an important way for visitors to engage with the everyday life of the neighborhoods, the existing supply has not yet been consistently leveraged as an experiential advantage.

Download:

• PNG
  larger image
• TIFF
  original image

Table 9. IPA evaluation results for tourism image perceptual elements.

https://doi.org/10.1371/journal.pone.0355218.t009

The differences were mainly reflected in Space-layer access conditions, Place-layer natural and ecological resources, and Setting- and Realm-related elements of everyday-life experience. In Longwei Pass and Ming-Qing Tonghai Old Town, C2 (natural ecological landscape) and A1 (location and transportation) were both located in Quadrant IV, reflecting a high-attention–low-performance mismatch among mountain–water resources, neighborhood accessibility, and the organization of actual visits. In Longwei Pass, C1 (traditional street-and-alley pattern), C4 (local folk life), and D1 (physical and mental experience) were all located in Quadrant I, indicating the perceptual advantage generated by the continuity of street-and-alley life. In Santai Mountain Historic Area, A1 (location and transportation), C2 (natural ecological landscape), C3 (historical and cultural relics), and D1 (physical and mental experience) were all located in Quadrant I, whereas B1 (food and snacks) and B2 (photography and social media check-ins) were located in Quadrant IV. This indicates that its public recreational spaces and historical, cultural, and educational nodes have basic appeal, while its consumption settings and social-sharing potential are not yet fully translated into perceptual advantages.

Overall, the perceptual differences among the three sites were to some extent associated with the degree of tourism development, namely Ming-Qing Tonghai Old Town > Longwei Pass > Santai Mountain Historic Area. However, this relationship does not follow a simple linear pattern; rather, it is jointly shaped by the organization of spatial resources, the embedding of everyday-life settings, the formation of place-based understanding, and the capacity to translate services into the visitor experience.

4.3.2. Differences in the perceptual structure of tourism image and their formation logic.

Based on the quadrant classification above, Fig 5(a) presents the overall distribution of HSSPR-coded perceptual elements across the three sites in the coordinate system of “attention salience–experiential performance”. The results show that high-attention elements within different HSSPR layers do not necessarily correspond to high experiential performance. This indicates that the tourism image of living HCNs depends not only on resource endowment or tourism development level, but also on how Space-layer access conditions, Setting-layer activities, and Place-layer resources are organized into accessible, understandable, and experienceable systems. The following analysis further explains the perceptual differences among the three cases based on their IPA distributions.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. IPA quadrant analysis of tourism image perceptual elements.

https://doi.org/10.1371/journal.pone.0355218.g005

The IPA results for Longwei Pass show a strong linkage among Place recognition, Setting participation, and Realm-level experience (Fig 5(b)). C1 (traditional street-and-alley pattern), C3 (historical and cultural relics), C4 (local folk life), and D1 (physical and mental experience) were all located in Quadrant I, indicating a stable perceptual association among historic pass remains, traditional streets and alleys, and the everyday marketplace life of old Xiaguan. Visitors can therefore move relatively easily from historical recognition to street-and-alley strolling and everyday-life experience. However, C2 (natural ecological landscape), B1 (food and snacks), B4 (commercial consumption), and A1 (location and transportation) remained in the priority improvement quadrant, suggesting that the Cangshan–Erhai setting, light-consumption formats, and neighborhood accessibility have not yet been fully transformed into stable experiential advantages. The key issue for Longwei Pass is therefore to improve transport access, heritage interpretation, and light-consumption organization while maintaining the authenticity of street-and-alley life.

The perceptual structure of the Santai Mountain Historic Area is more complex, with advantages concentrated in Space-, Place-, and Realm-layer elements but weaknesses remaining in Setting-layer consumption and social-sharing elements (Fig 5(c)). A1 (location and transportation), C2 (natural ecological landscape), C3 (historical and cultural relics), and D1 (physical and mental experience) were all located in Quadrant I, indicating that its mountainous environment and religious, cultural, and educational nodes jointly constitute its basic appeal. However, B1 (food and snacks) and B2 (photography and social media check-ins) fell into the priority improvement quadrant, while B4 (commercial consumption), C1 (traditional street-and-alley pattern), and D3 (memory and spiritual sustenance) were mainly located in the low-priority quadrant. D2 (place identity) showed relatively high performance but limited attention. These results suggest that although visitors can recognize its natural environment and historical nodes, they have difficulty forming continuous consumption experiences, social-media-based memories, and an integrated understanding of place. This pattern is related to its limited tourism development, the point-like distribution of resources, the relatively clear access boundaries of religious, cultural, and educational spaces, and the fact that community life has not yet been effectively organized into tourism narratives.

Ming-Qing Tonghai Old Town has a relatively strong Place-layer resource foundation, but its transformation into continuous Setting- and Realm-level experiences remains insufficient (Fig 5(d)). C3 (historical and cultural relics) and D4 (revisit and recommendation intention) were located in Quadrant I, suggesting that Confucian Temple culture, the ancient-town pattern, and local life are highly recognizable and can stimulate a certain level of recommendation intention. However, C2 (natural ecological landscape), as the perceptual element with the highest attention salience, did not fully meet visitor expectations regarding experiential performance. A1 (location and transportation), B1 (food and snacks), and B4 (commercial consumption) were also located in the priority improvement quadrant, revealing deficiencies in access, consumption, and service support. C1 (traditional street-and-alley pattern) and C4 (local folk life) had not yet developed prominent experiential advantages. Although Xiushan Mountain, the old town, and Qilu Lake form a relatively complete mountain–water, cultural, and educational structure, visitor-route connections, transport transfers, and service capacity still limit the transformation of the overall resource system into a continuous visitor experience.