Next Article in Journal
Nitric Oxide-Mediated Regulation of Cadmium Stress in Lettuce Varieties: Implications for Sustainable Food Safety and Crop Production
Previous Article in Journal
The Impact of Suppliers’ High-Quality Interactions and Financial Constraints on Downstream Firms’ Operational Resilience: The Moderating Effect of Suppliers’ ESG Performance
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 18
Issue 4
10.3390/su18041936
sustainability-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Mechanisms Underlying the Effects of Recreation Services on Tourist Satisfaction in Forest Parks: A Case Study of the Yangtze River Delta, China

by
Caijie Chen
†,
Weilin Zhao
† and
Bing Zhao
*
College of Landscape Architecture, Nanjing Forestry University, Nanjing 210018, China
*
Author to whom correspondence should be addressed.
†
These authors contributed equally to this work.
Sustainability 2026, 18(4), 1936; https://doi.org/10.3390/su18041936
Submission received: 14 January 2026 / Revised: 10 February 2026 / Accepted: 11 February 2026 / Published: 13 February 2026
Browse Figures
Versions Notes

Abstract

Recreation service evaluation systems are critical to forest park management; however, existing frameworks often emphasize static infrastructure while overlooking tourists’ dynamic perceptions and sentiments. This study develops a comprehensive recreation service evaluation framework by integrating objective geospatial indicators with social media-based tourist feedback. A total of 67 forest parks in the Yangtze River Delta were selected as the study area. Descriptive statistics and spatial autocorrelation analyses, including Global and Local Moran’s I, were applied to the statistical properties and spatial patterns of recreation service indicators, tourist comments, and sentiment. In addition, an Extreme Gradient Boosting model with SHAP interpretation was employed to identify key recreation service indicators and explore their non-linear effects on tourist participation and sentiment. The results reveal significant positive spatial clustering of tourist comments and sentiment. Service-related and safety indicators play a dominant role in shaping tourist participation and sentiment patterns, whereas ecological indicators show weaker direct effects. Several indicators exhibit clear non-linear and threshold effects. Overall, recreation service experiences in forest parks are primarily driven by accessibility, service facilities, and safety assurance, and the proposed framework enhances dynamic interpretability and supports the optimization of recreation services.

1. Introduction

China’s forest area is approximately 2.2 million square kilometers, accounting for around 23% of the country’s total land area [1,2,3]. These forests provide very important sites for ecotourism, and this process has led to the emergence of a specific tourism form known as forest tourism [4,5,6]. In recent years, the Chinese government has implemented a series of policies for forest resource protection and promoted the development of forest tourism, making it a favored travel option among the majority of tourists [7,8,9,10].
As an important venue for forest tourism, a forest park is defined as a specific forested area with a sound ecosystem and significant tourism development value [11,12,13,14]. In addition to their most basic recreational value, China’s forest parks have received protection and restoration from the Chinese government and often have a certain landscape character that can provide some artistic value [15,16,17,18]. As a result, forest parks have increasingly attracted scholarly attention. Some studies have examined factors influencing forest tourism satisfaction using questionnaire surveys and structural equation modeling and have found that a good ecological environment and rich forest resources can attract tourists [7,19].
Meanwhile, with the development of forest park tourism, the tourism evaluation system for forest parks has gradually revealed its drawbacks, such as insufficient environmental monitoring of forest parks [20,21,22], incomplete tourism evaluation systems and so on [13,23,24]. Therefore, we referred to the solutions of other parks and found that most parks use remote sensing technology to monitor and assess forest activity indicators [25,26], such as monitoring the role of buffer zones in limiting deforestation in Honduras’ Cerro Azul Meambar National Park and Nepal’s Chitwan National Park using remote sensing technology [26,27], evaluation of forest cover change using remote sensing techniques and landscape metrics in Moncayo Natural Park (Spain), and modeling the spatial variation in urban park ecological properties using remote sensing data [28]. In remote sensing, although large-scale geographic information can be acquired, limitations exist in capturing details, and as resolution decreases, accuracy will decline [29,30].
In recent years, the rise of social media has opened new avenues for park tourism data collection. Tourists can provide crowdsourced contextual description information of forest parks through social media, such as text and images [31,32]. This compensates for the one-sidedness of remote sensing data and the abstractness of geospatial data. Earlier research has confirmed the reliability of social media data in tourism, as well as the possibility of high spatial and temporal resolution, which is beneficial for nature-based tourism research [33,34]. By observing tourists’ experiences and perceptions across spatial and temporal scales, these data provide an empirical basis for examining service quality and satisfaction-related theoretical constructs. Expectancy–Disconfirmation Theory explains how gaps between tourists’ expectations and post-visit experiences shape satisfaction outcomes [35,36,37,38]; SERVQUAL conceptualizes service quality through tangible infrastructure, accessibility, and functional performance [39,40,41,42]; Experience Economy theory emphasizes the role of environmental and experiential attributes in enhancing recreational value and satisfaction [43,44,45,46]. Nevertheless, despite well-established theoretical foundations, empirical applications of these theories remain limited, as most existing studies focus on urban parks rather than forest parks [47,48].
Multimodal data can address this deficiency and has already gained development in other fields, such as Suzuki et al., using geospatial multimodal data for forest cover classification [49]. Lamahewage et al. estimated aboveground biomass using multimodal remote sensing observations and machine learning in temperate mixed forests [50]. Zheng et al. synergized multi-modal data for mapping tree species distribution [51]. Ling et al. monitored canopy height in the Hainan tropical rainforest via machine learning and multi-modal data fusion [52]. However, this method was not used to improve the tourism evaluation system of forest parks.
Therefore, to make up for the aforementioned research gaps, this paper primarily explores the following three questions: (1) Do different types of recreation service indicators in forest parks exhibit significant differences and imbalances in their quantity levels and spatial distribution? (2) How do different types of recreation service indicators influence tourist participation and sentiment responses, and do these effects exhibit non-linear or threshold characteristics? (3) How can objective geospatial data be integrated with social-media-comment-based sentiment information to construct a more interpretable and management-oriented evaluation framework for recreation services in forest parks?
To address the above issues, this study takes the Yangtze River Delta as a case study to examine the mechanisms through which recreation services influence tourist satisfaction in forest parks from a multimodal data perspective. By integrating remote sensing data, geospatial information, and social media reviews, a comprehensive evaluation framework for recreation services and tourist satisfaction is established. Correlation analysis and interpretable machine learning methods are employed to investigate the relationships between recreation service indicators and both the volume and sentiment characteristics of tourist reviews. By identifying heterogeneity and threshold effects, this study aims to enhance understanding of how recreation services shape tourist satisfaction and to provide empirical references for differentiated management strategies and service optimization in forest parks.

2. Materials and Methods

2.1. Study Area

Endowed with favorable geographical and climatic conditions, the Yangtze River Delta Urban Agglomeration supports abundant forest resources and has experienced the progressive development of a diverse system of forest parks. These parks integrate rich natural and cultural landscapes while accommodating high levels of tourism activity, generating extensive and multidimensional tourist evaluation data. Accordingly, this study focuses on the Yangtze River Delta Urban Agglomeration and selects 67 forest parks as research samples (Figure 1). Spanning major cities and counties across Shanghai, Jiangsu, Zhejiang, and Anhui provinces, the sample includes national, provincial (municipal), and county-level forest parks, ensuring broad spatial coverage and strong representativeness.

2.2. Materials and Variables

The tourism evaluation system of forest parks is of great significance for the tourism development of forest parks. Building on the indicator characteristics proposed by Yang [53] and Jalilian [54], this study improves the tourism evaluation system of forest parks by incorporating geospatial data and social media reviews.

2.2.1. Tourist Satisfaction Variables

Tourist satisfaction variables were primarily derived from social media reviews, which capture tourists’ experiential evaluations and sentiment responses to forest parks. Social media data were extracted from Ctrip, the largest online travel service platform in mainland China [55]. Accordingly, these reviews allow for the theoretical operationalization of tourist satisfaction based on visitors’ post-experience evaluations.
As expectancy–disconfirmation theory conceptualizes tourist satisfaction as the discrepancy between prior expectations and actual recreational experiences. In this study, tourist satisfaction is assessed by jointly considering emotional evaluations (sentiment polarity) and behavioral engagement (review quantity), which together provide satisfaction outcomes following experience evaluation. Accordingly, the tourist interviews dimension includes tourism service review quantity (TSRQ) and total review quantity (TRQ), the tourist sentiment dimension comprises four indicators: total sentiment (TE), negative sentiment (NEG), neutral sentiment (NE), and positive sentiment (PE). Collectively, these six indicators constitute a multidimensional representation of tourist satisfaction [56,57,58]. To further characterize the content structure of tourist comments, TSRQ was decomposed into six category-specific review ratios, including recreation service review ratio (RSRR), science and education service review ratio (SERR), public facility review ratio (PFRR), transportation review ratio (TRR), safety service review ratio (SRR), and landscape review ratio (LRR), which reflect tourists’ relative attention to different service dimensions.

2.2.2. Recreation Service Evaluation Variables

This study integrates the recreational service demands of forest parks with visitors’ experiential characteristics and constructs an evaluation framework comprising six dimensions: infrastructure and accessibility; tourism services and supporting facilities; safety and operational assurance; ecological environmental quality; education, culture, and experiential services; and tourist satisfaction. Based on this framework, 15 indicators derived from remote sensing and geospatial data were selected (Table 1), including scientific and educational diversity (SED), scientific and educational facilities (SEF), accommodation facilities (AF), landscape ancillary facilities (LAF), shopping facilities (SF), transportation facilities (TF), road network density (RND), bus stops (BS), parking lots (PL), public security agencies (PSA), medical institutions (MI), fire departments (FD), infrastructure update time (IUT), vegetation diversity (VDEN), and vegetation density (VDIV). Area of interest (AOI) boundary data were used to delineate the spatial extent of each forest park, and a 1 km buffer zone was subsequently constructed around each AOI to quantify surrounding service facilities and environmental conditions using point of interest (POI) data, normalized difference vegetation index (NDVI) data, and road network data.
The selection of indicators was guided by their relevance to visitors’ core demands for travel convenience, service completeness, and safety, as well as their preferences for ecological experiences and cultural and educational activities. By relying on objective remote sensing and geospatial data, the selected indicators provide broad spatial coverage and reliable measurement, offering a robust basis for subsequent spatial analysis and comparative evaluation across forest parks.

2.3. Data Preprocessing

2.3.1. Geographic Information Data Processing

For spatial visualization, the natural breaks (Jenks) classification method was used to divide each indicator into five classes based on its value distribution. This classification was applied solely for map visualization and spatial pattern exploration, while all statistical analyses including correlation analysis, XGBoost regression, and SHAP-based interpretation were conducted using the original continuous variables. The indicators analyzed include scientific and educational diversity, scientific and educational facilities, accommodation facilities, landscape ancillary facilities, shopping facilities, transportation facilities, road network density, bus stops, parking lots, public security agencies, medical institutions, fire departments, infrastructure update time, vegetation diversity, and vegetation density.
To characterize the surrounding recreational service environment influencing visitor experiences, 1 km and 10 km buffer zones were constructed around each forest park [59,60]. This buffer captured nearby communities, transportation nodes, and supporting facilities, allowing the analysis to incorporate both park-level characteristics and external environmental contexts and providing a robust basis for subsequent spatial analyses.

2.3.2. Social Media Data Processing

For the preprocessing of social media data, Octopus Collector was selected as the data acquisition tool, and non-intrusive web crawling technology was employed to retrieve tourist review data of forest parks from Ctrip. The field extraction rules were iteratively optimized and validated, with a focus on capturing six categories of key information: park names were obtained via semantic parsing of HTML title tags; scenic area levels were extracted from the attribute values of 5A/4A badge icons; user ratings were recorded with a precision of one decimal place (e.g., 4.8 points); review texts were fully captured for all visible entries using a cyclic list mode; publication dates were standardized into the date format (YYYY-MM-DD) through regular expression matching; and user geographic locations were extracted as provincial-level administrative units from tagged descriptions.
In response to the pagination logic of the Ctrip platform, the system was configured with an automatic page-turning function, and data collection continued until all reviews published by 31 December 2024 were exhausted, resulting in an original dataset of 186,000 reviews. The data cleaning phase adopted a multi-level quality control strategy, encompassing the following core steps: system-generated default reviews, duplicate reviews, short texts (fewer than five characters), and irregular web expressions (e.g., emojis, non-Chinese characters, and informal internet slang) were identified and removed through multi-step preprocessing, including text normalization and dictionary-based filtering. Subsequently, the cleaned texts were segmented into sentences using regular expressions and tokenized with the jieba library enhanced by domain-specific dictionaries, followed by stop-word removal using multiple authoritative Chinese stop-word lists to produce a high-quality corpus for subsequent analysis.
Sentiment analysis uses natural language processing techniques to quantify tourists’ sentiment tendencies in online reviews by transforming unstructured text into measurable sentiment polarity and intensity. In this study, cleaned and standardized tourist reviews were analyzed using the Baidu Natural Language Processing (NLP) platform, which applies deep learning models (ERNIE and BiLSTM) to output sentiment categories and positive sentiment probabilities ranging from 0 to 1. To ensure construct validity, we manually coded a random 2% sample of the reviews (n = 372) and found substantial agreement with the Baidu NLP outputs, as measured by Cohen’s Kappa (Kappa = 0.78), which assesses inter-coder reliability by accounting for agreement occurring by chance [61,62]. At the park level, the average positive sentiment probability was used as the key dependent variable to examine how recreational service indicators influence tourists’ sentiment experiences and overall satisfaction. Figure 2 shows the processing of recreational services and tourist satisfaction and their contributions to the evaluation system.

2.4. Analysis Approach

2.4.1. Spatial Pattern Analysis

Prior to spatial analysis, descriptive statistics were computed to summarize the basic distributional characteristics of recreational service indicators across forest parks in the study area. For each indicator, the minimum, maximum, mean, standard deviation, and median were calculated to characterize central tendency and dispersion, thereby revealing overall variability and differences in recreational service provision among forest parks.
To examine spatial distribution patterns and spatial dependence, spatial autocorrelation analysis was performed using both Global Moran’s I and Local Moran’s I. Global Moran’s I was applied to assess whether forest parks with similar recreational service levels exhibited significant spatial clustering or dispersion across the region, with statistical significance evaluated using Z-scores and p-values. Based on the global results, Local Moran’s I was further used to identify local clustering and outlier patterns, classifying forest parks into High–High and Low–Low clusters as well as High–Low and Low–High outliers, thereby revealing spatial heterogeneity and potential hotspot and cold spot areas of recreational services.

2.4.2. Correlation Analysis

The Pearson product–moment correlation coefficient was employed to quantify the strength and direction of linear associations between two sets of continuous variables, specifically among the 15 recreational service indicators. The correlation coefficient r ranges from −1 to 1, where r > 0 indicates a positive correlation, r < 0 indicates a negative correlation, and values of |r| closer to 1 represent stronger linear relationships.
In addition, the Mantel test was employed to assess the overall association between two multivariate datasets while accounting for their matrix structure. The statistical significance of the observed association was evaluated using the p-value of the test statistic.
Finally, the Pearson correlation coefficients and Mantel test results were integrated into a heat map (hot plot) for visualization. In the plot, cell colors (ranging from green to pink) represent the sign and magnitude of the Pearson correlation coefficients, with numerical values indicating the exact coefficients. Line types (solid for positive associations and dashed for negative associations) and colors (pink for p < 0.01; green for 0.01 ≤ p < 0.05; white for p ≥ 0.05) correspond to the direction and significance levels of the Mantel test, providing an intuitive representation of the relationships between the two variable sets.

2.4.3. Extreme Gradient Boosting (XGBoost) Model

Proposed by Chentianqi, XGBoost is a high-performance gradient boosting algorithm [63]. Its core value lies in its ability to quantify and rank the relative importance of indicators from complex data.
The core analytical procedure was as follows. First, label encoding was applied to categorical variables in the recreation service indicator dataset, while missing values were retained to leverage XGBoost’s built-in missing value handling mechanism. The dataset was then randomly split into training and testing subsets at a ratio of 8:2. Subsequently, an XGBoost regression model was initialized to predict continuous outcome variables (i.e., tourist review quantity and sentiment indicators). A baseline model was trained and then optimized through hyper-parameter tuning. The final model achieved satisfactory predictive performance, with an R2 value of 0.51 on the test set.
Based on the trained model, extract the gain importance to quantify the contribution of each recreational service indicator to satisfaction prediction, integrate SHAP value analysis to interpret the directional impact of indicators, and generate an importance ranking chart to clearly label the top 5 core indicators, providing data support for the management optimization of forest parks.

2.4.4. Explanation Based on Shapley Methods

SHAP (Shapley Additive Explanations) is an explainable machine learning framework grounded in cooperative game theory, proposed by Lundberg and Lee [64]. Its core idea is to interpret the predicted value of the model as the sum of the marginal contributions of each feature to the prediction result, and quantify the feature importance through the Shapley value.
In this study, SHAP analysis was conducted to interpret the nonlinear relationships captured by the XGBoost model. First, the trained XGBoost regression model and its corresponding input feature set were used as the basis for model interpretation.
Subsequently, the SHAP library in Python 3.13.0 was employed to construct an appropriate explainer, and SHAP values were calculated for each recreational service indicator to quantify their marginal contributions to model predictions. For each indicator, SHAP dependence plots were generated with feature values on the x-axis and corresponding SHAP values on the y-axis. A quadratic fitted curve was applied to illustrate nonlinear trends and potential threshold effects. Finally, Spearman’s rank correlation coefficients were computed between indicator values and their SHAP values to quantify the strength and direction of feature effects and assess their statistical significance.

2.4.5. Robustness Analysis of Model Specification

To assess the robustness of the results, a cross-model validation strategy was first implemented using two machine learning models with different learning mechanisms, namely CatBoost and Random Forest. Model performance was evaluated using R2, MAE, MSE, and RMSE, and the consistency of key predictors was examined through feature importance rankings. Second, to mitigate potential endogeneity concerns, an ordinary least squares (OLS) regression was conducted by interchanging the independent and dependent variables, and the stability of the results was assessed through comparison of the estimated coefficients.
Finally, to further evaluate the robustness of the multi-spatial scale, an alternative 10 km buffer was constructed for each forest park, and feature importance analyses were recalculated.

3. Results

3.1. Analysis of Descriptive Statistical Results of Recreation Service Indicators and Tourist Satisfaction

3.1.1. Descriptive Statistics and Spatial Distribution of Tourist Satisfaction

Although indicators are classified into five levels for visualization, all quantitative analyses are based on original continuous values (Figure 3). Visitor comment and sentiment indicators exhibit pronounced spatial clustering across forest parks in the Yangtze River Delta. Figure 3a illustrates the proportions of comments associated with different service categories using different colors (e.g., recreational services, science education facilities, public facilities). Overall, among the 67 forest parks in the Yangtze River Delta, comments on recreational services and landscape features each account for more than half of the total comments, indicating that visitors pay greater attention to landscape facilities and recreational experiences. In addition, eastern cities exhibit a greater diversity of comment categories, whereas western cities show relatively more homogeneous comment types.
Figure 3b presents the proportions of positive, neutral, and negative tourist sentiments. In summary, tourist sentiment in the 67 forest parks of the Yangtze River Delta is predominantly positive, followed by neutral sentiment, with negative sentiment accounting for a relatively small share. Spatial variation in sentiment distribution is limited, suggesting a generally consistent pattern of sentiment feedback on tourist experiences across the Yangtze River Delta. Forest parks located in eastern cities also show a relatively higher proportion of positive sentiment.

3.1.2. Descriptive Statistics and Spatial Distribution of Recreation Service Indicators

Recreation service indicators exhibit distinct spatial patterns and statistical characteristics and can be grouped into four aspects (Figure 4; Table 2): service facilities, review and sentiment, ecological environment, and basic support.
The service facility aspect, including accommodation facilities, parking lots, and bus stops, shows pronounced high-value clusters in southern Jiangsu and northern Zhejiang. Although mean values are moderate, large standard deviations and medians substantially lower than the means (e.g., accommodation facilities: mean = 121.30, SD = 295.85) indicate a highly uneven distribution, with facilities concentrated in a limited number of areas characterized by high population density and tourist intensity.
Indicators within the review and sentiment aspect exhibit high-value clusters in popular scenic areas, largely overlapping with service facility hotspots. These indicators are characterized by high means and extremely large standard deviations (e.g., total review quantity: mean = 1350.64, SD = 1396.02), reflecting strong dependence on visitor activity. Specifically, a small number of highly frequented hotspots dominate the overall spatial pattern, thereby inflating average values.
The ecological environment aspect, such as vegetation diversity and vegetation density, displays a relatively homogeneous spatial distribution with weak clustering. Low means and small standard deviations (e.g., vegetation density: mean = 0.61, SD = 0.13) suggest limited human disturbance and a consistent ecological baseline across forest parks.
In contrast, the basic support aspect (e.g., PSA, MI, and FD) shows dispersed high-value points centered on towns rather than concentrated clusters. These indicators exhibit low means and moderate variability (e.g., public security agencies: mean = 34.20, SD = 58.46), indicating generally adequate coverage of public support facilities, with variability mainly driven by differences in town size.

3.1.3. Spatial Autocorrelation Patterns of Tourist Satisfaction

Figure 5 illustrates the spatial autocorrelation patterns of visitor comments and sentiment indicators based on Moran’s I analysis. The Global Moran’s I results show that both review quantity and sentiment indicators exhibit significant positive spatial autocorrelation, indicating that forest parks with similar levels of visitor engagement and sentiment expression in the Yangtze River Delta are not randomly distributed but instead display clear spatial clustering.
The Local Moran’s I analysis further reveals distinct spatial clusters and outlier patterns. High–High (HH) clusters of comments and positive sentiment are mainly concentrated in forest parks located within and around core metropolitan areas, such as Shanghai, southern Jiangsu, and northern Zhejiang, reflecting strong spatial spillover effects of tourism popularity and visitor engagement. These areas are characterized by high visitor flows and diversified recreational services, which jointly enhance comment intensity and positive sentiment expression. In contrast, Low–Low (LL) clusters are primarily distributed in peripheral or less accessible forest parks, such as Tongling Mountain National Forest Park and Da shushan National Forest Park, corresponding to relatively low levels of visitor participation and limited sentiment feedback.

3.2. Analysis of the Impact Mechanisms of Recreational Service Indicators on Tourist Satisfaction and Key Indicators

3.2.1. Results of the Correlation Analysis of Recreational Service Indicators

Based on the Pearson correlation and Mantel test heatmap (Figure 6), a clear and systematic association is observed between recreational service indicators and tourist satisfaction. Most recreational service facility indicators (e.g., AF, LAF, SF, BS, PL, PSA, MI, and FD) exhibit significant positive correlations with positive sentiment (PE) and total sentiment (TE), with the majority of Mantel test results reaching statistical significance (p < 0.05 or p < 0.01). This indicates that higher levels of recreational service provision are strongly associated with more positive sentiment experiences among tourists. In contrast, correlations between recreational service indicators and negative sentiment (NEG) are generally weak and mostly insignificant, suggesting that improvements in recreational services primarily enhance positive sentiment rather than directly reducing negative sentiment. Neutral sentiment (NE) shows no stable or significant relationship with service indicators, indicating limited sensitivity to changes in recreational services.
In terms of tourist behavior, total review quantity (TRQ) is significantly and positively correlated with several recreational service indicators, implying that well-developed service facilities not only improve sentiment responses but also stimulate visitor engagement and review activity.
Correlations among recreational service indicators reveal strong internal linkages, reflecting coordinated development and functional complementarity among facilities. Conversely, ecological indicators show negative correlations with construction-related variables, highlighting trade-offs between development intensity and ecological quality. Collectively, recreational service indicators demonstrate strong internal coherence and a pronounced positive association with tourist satisfaction.

3.2.2. The Importance of Various Recreational Service Indicators for Review Volume and Tourist Sentiment

Figure 6 presents the results of the XGBoost model, which evaluates the importance of recreational service indicators from two perspectives: social media comment popularity and tourist sentiment. Figure 7a illustrates the contributions of individual indicators to comment popularity. Scientific and educational facilities emerge as the most influential factor, with a feature importance value of 0.204, accounting for 27.8% of the top five indicators. Parking lots, medical institutions, and public security agencies also exert significant effects on the prediction results. In contrast, indicators such as vegetation density, infrastructure update time, and accommodation facilities exhibit importance values below 0.030, indicating a relatively weak contribution to review popularity.
Figure 7b shows that scientific and educational facilities and parking lots are the main drivers of tourist sentiment, representing 41.5% and 26.4% of the total importance, respectively. Together, they account for over 67% of the weight. Safety-related indicators, including medical at 14.2%, fire at 9.4%, and security at 8.5%, also show substantial influence. The variation trends of these results are consistent across the 1 km and 10 km buffers.

3.2.3. Threshold Effect Analysis of Recreational Service Indicators on Review Popularity in Forest Parks

Based on the SHAP dependence analysis with tourist review popularity as the dependent variable (Figure 8 and Figure A1), recreational service indicators exhibit four response patterns, including unimodal, U-shaped, linearly increasing, and linearly decreasing.
Several core indicators display pronounced threshold effects, reflecting tourists’ sensitivity to functional adequacy rather than absolute service abundance. Parking lots show a clear unimodal pattern, with review popularity increasing rapidly when parking supply is insufficient, as parking inconvenience directly constrains travel behavior and experience quality. Review popularity peaks at approximately (PL = 466), beyond which additional parking provides limited perceptible benefits and gradually reduces tourists’ motivation to review. Similar threshold-dependent responses are observed for landscape ancillary facilities (LAF = 274) and scientific and educational diversity (SED = 3.73), indicating that moderate provision enhances experiential richness and interaction opportunities, while excessive provision does not further stimulate review activity. Fire departments show a linearly increasing effect on review popularity, as improved emergency response capacity continuously enhances tourists’ perceived safety and trust, thereby steadily encouraging engagement and review behavior without clear saturation effects.
Infrastructure condition also exerts a consistent influence on tourist engagement. Infrastructure update time shows a linear negative relationship with review popularity, indicating that as facilities age, declining functionality and visual quality continuously weaken tourists’ willingness to engage and post reviews, underscoring the importance of timely infrastructure renewal. By contrast, variables such as medical institutions (MI) and other secondary service indicators exhibit relatively flat SHAP response curves, suggesting limited marginal influence on tourist review popularity within the observed range.
Taken together, the SHAP results demonstrate that tourist engagement is most strongly driven by service indicators with explicit threshold effects, particularly parking facilities, public safety infrastructure, and infrastructure update cycles, highlighting that meeting tourists’ basic functional expectations is more effective than the unlimited expansion of service supply in stimulating review popularity.

3.2.4. Threshold Effect Analysis of Recreational Service Indicators on Tourist Sentiment in Forest Parks

Based on the SHAP dependence analysis with tourist sentiment as the dependent variable (Figure 9 and Figure A2), recreational service indicators exhibit four response patterns: unimodal, U-shaped, linearly increasing, and linearly decreasing.
Among the indicators, parking lots and scientific and educational diversity (SED) exhibit pronounced unimodal patterns with clear threshold effects. Prior to their respective thresholds (PL = 479, SED = 6.11, FD = 33.3), increases in parking supply and educational diversity effectively alleviate access constraints and enrich experiential content, thereby continuously enhancing positive sentiment responses. However, once these thresholds are exceeded, excessive facility provision tends to encroach upon recreational or ecological space and reduce experiential coherence, leading to a gradual decline in tourist sentiment. These results highlight that tourist sentiments are most sensitive to the adequacy rather than the abundance of core service facilities.
Other indicators, including accommodation facilities, fire departments, landscape ancillary facilities, public security institutions, transportation-related indicators, and shopping facilities, also exhibit nonlinear or linear response patterns but with comparatively weaker effects. Their influences mainly reflect general functional matching and environmental coordination and do not dominate changes in tourist sentiment responses.
Overall, the SHAP results indicate that tourist sentiment is primarily driven by key indicators with explicit threshold or monotonic effects, particularly parking facilities, scientific and educational diversity, and infrastructure update time, rather than by uniformly increasing all types of recreational service provision.

3.3. Robustness Analysis Result

The robustness of the results was confirmed through three tests. First, cross-model validation showed similar performance across CatBoost (R2 = 0.43), Random Forest (R2 = 0.45), and XGBoost (R2 = 0.51), with the top three predictors remaining identical. Second, in the ordinary least squares (OLS) reciprocal test, interchanging the dependent and independent variables yielded no statistically significant results, addressing potential endogeneity. Finally, re-running the analysis with a 10 km buffer showed that key variable trends remained consistent, confirming the robustness and spatial stability of the identified threshold effects.

4. Discussion

The findings of this study provide important insights into the mechanisms through which recreation services in forest parks influence tourist reviews and sentiment responses. The spatial clustering of service facilities and tourist sentiment in developed cities and popular tourist destinations indicates that accessibility and tourism facilities play a critical role in shaping visitor experiences. This pattern is consistent with previous studies that emphasize the importance of accessibility and recreational service infrastructure in leisure spaces [65,66,67,68,69,70], while also highlighting the uneven distribution of recreational service resources among forest parks.
The results further indicate that, unlike urban settings where ecological condition varieties (e.g., vegetation diversity, vegetation density) strongly shape visitor perceptions [71,72,73], forest parks generally exhibit relatively high and homogeneous ecological conditions [74,75]. Therefore, ecological conditions are more likely to be perceived as baseline expectations, while differentiated recreational services become more salient indicators of tourist satisfaction [76]. Although the existing literature frequently underscores the role of landscape quality and ecological integrity in forest recreation [77,78,79,80], the findings of this study suggest that improvements in ecological conditions alone may not substantially enhance tourist satisfaction when basic service and safety needs are not adequately met. In addition, the identification of nonlinear and threshold effects extends previous research that commonly assumes linear relationships between recreation service provision and visitor satisfaction, indicating that excessive facility provision may undermine tourist satisfaction by increasing crowding or diminishing the sense of naturalness.
From a practical perspective, the findings highlight that forest parks need to adopt differentiated management strategies to improve service provision, accessibility and safety in a targeted manner, rather than pursuing uniform infrastructure expansion. Methodologically, the integration of spatial analysis with sentiment information derived from social media comments demonstrates a dynamic framework for evaluating recreation service levels, providing a valuable complement to traditional static assessment methods and supporting more adaptive forest park management.
Despite these contributions, several limitations should be acknowledged. First, social media data may introduce demographic bias by primarily representing younger visitors, while other visitor groups (e.g., elderly, non-users of technological applications, etc.) may be underrepresented [81,82,83]. Second, the findings are derived from the Yangtze River Delta, and their applicability to regions with different social, geographical, and climatic characteristics may require further validation. Third, the identified SHAP threshold effects are contingent on the model configuration and the specific spatial–temporal characteristics of the dataset, which may limit their direct generalizability. Finally, given the cross-sectional data, the managerial and policy implications are exploratory rather than causal. However, these limitations mainly relate to data representation and analytical scope and do not affect the robustness of the main findings. Future studies could integrate social media data with survey and panel data, extend the analysis to multiple regions, and employ geographically weighted approaches and random forest models to account for spatial dependence.

5. Conclusions

This study integrates geospatial indicators with social media-based data to systematically evaluate recreation services in forest parks.
The results reveal pronounced spatial heterogeneity in recreation service indicators, with service facilities and tourist sentiment highly concentrated in developed cities and popular scenic areas, while ecological indicators remain relatively homogeneous across forest parks. Service-supporting and safety-related facilities are identified as the primary drivers of tourist participation and sentiment responses, whereas ecological indicators exhibit relatively limited direct effects. The proposed framework demonstrates that combining spatial analysis with sentiment data enables a more dynamic and interpretable evaluation of forest park recreation services.
Based on the research findings, policymakers should implement differentiated management strategies according to the level of tourist satisfaction in forest parks. For parks with lower satisfaction levels, greater and more appropriate investment should be directed toward recreational service facilities and accessibility to address deficiencies in basic services. In contrast, parks with higher satisfaction levels should prioritize the timely renewal and functional optimization of infrastructure to continuously enhance the quality of recreational services. Meanwhile, forest parks should be integrated into regional transportation and public service planning, with strengthened coordination of surrounding public services, and the application of multi-source data in park management and policy formulation should be promoted to achieve scientific governance and high-quality development of forest parks. Overall, this study provides empirical references for optimizing recreation services and supports more balanced and sustainable forest park management.

Author Contributions

Conceptualization, software, validation, investigation, resources, data curation, C.C.; Methodology, formal analysis, data curation, writing—original draft preparation, visualization, W.Z.; writing—review and editing, funding acquisition, B.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Natural Science Foundation of China (No. 32471941).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable. This study used publicly available social media data and did not involve direct interaction with human participants. No identifiable personal information was collected.

Data Availability Statement

The raw data supporting the conclusions of this article will be made available by the authors on request.

Acknowledgments

We thank the editors and reviewers of the journal for their valuable suggestions for improving the manuscript.

Conflicts of Interest

The authors have no conflicts of interest to declare that are relevant to the content of this article.

Appendix A

Figure A1 supplements the conclusions presented in Section 3.2.3. The relatively high counts of parking and accommodation facilities result from the use of a 1km buffer around forest parks, which incorporates nearby service facilities that substantially influence visitor convenience and engagement.
Sustainability 18 01936 g0a1
Figure A1. Correlation Analysis of Recreational Service Indicators and Comment Popularity.
Figure A1. Correlation Analysis of Recreational Service Indicators and Comment Popularity.
Sustainability 18 01936 g0a1
This Figure A2 is used to supplement the conclusions drawn in Section 3.2.4 of the article.
Sustainability 18 01936 g0a2
Figure A2. Correlation Analysis of Recreational Service Indicators and tourist sentiment.
Figure A2. Correlation Analysis of Recreational Service Indicators and tourist sentiment.
Sustainability 18 01936 g0a2

References

  1. Achard, F.; Hansen, M.C. Global Forest Monitoring from Earth Observation; CRC Press: Boca Raton, FL, USA, 2012. [Google Scholar]
  2. Zhang, D. China’s forest expansion in the last three plus decades: Why and how? For. Policy Econ. 2019, 98, 75–81. [Google Scholar] [CrossRef]
  3. Sun, W.; Liu, X. Review on carbon storage estimation of forest ecosystem and applications in China. For. Ecosyst. 2019, 7, 4. [Google Scholar] [CrossRef]
  4. Wang, L.; Zhao, H.; Wu, W.; Song, W.; Zhou, Q.; Ye, Y. Time-Varying Evolution and Impact Analysis of Forest Tourism Carbon Emissions and Forest Park Carbon Sinks in China. Forests 2024, 15, 1517. [Google Scholar] [CrossRef]
  5. Sgroi, F. Forest resources and sustainable tourism, a combination for the resilience of the landscape and development of mountain areas. Sci. Total Environ. 2020, 736, 139539. [Google Scholar] [CrossRef]
  6. Tampakis, S.; Andrea, V.; Karanikola, P.; Pailas, I. The growth of mountain tourism in a traditional forest area of Greece. Forests 2019, 10, 1022. [Google Scholar] [CrossRef]
  7. Yang, Z. Analysis of Influencing Factors of Tourist Satisfaction in Fuzhou Forest Park based on Structural Equation Model. Front. Bus. Econ. Manag. 2023, 12, 202–203. [Google Scholar] [CrossRef]
  8. Zhang, L.; Wu, C.; Hao, Y. Effect of the development level of facilities for forest tourism on tourists’ willingness to visit urban forest parks. Forests 2022, 13, 1005. [Google Scholar] [CrossRef]
  9. Daxin, Y.; Ziwei, J.; Zhexin, W.; Hede, G. Development and Research of Forest Tourism from the Perspective of Forest Experience. J. Landsc. Res. 2016, 8, 116. [Google Scholar]
  10. Sun, Q.; Zhang, N.; Liu, Z.; Liao, B. Tourism resources and carrying capacity of scenic tourism areas based on forest ecological environment. South. For. J. For. Sci. 2020, 82, 10–14. [Google Scholar] [CrossRef]
  11. Hu, J.; Wu, Y.; Irfan, M.; Hu, M. Has the ecological civilization pilot promoted the transformation of industrial structure in China? Ecol. Indic. 2023, 155, 111053. [Google Scholar] [CrossRef]
  12. Kang, N. Assessing Tourism Carrying Capacity Based on Visitors’ Experience Utility: A Case Study of Xian-Ren-Tai National Forest Park, China. Forests 2023, 14, 1694. [Google Scholar] [CrossRef]
  13. Lu, J.; Chen, H. Dynamic Evaluation and Forecasting Analysis of Touristic Ecological Carrying Capacity of Forest Parks in China. Forests 2023, 15, 38. [Google Scholar] [CrossRef]
  14. Golos, P.; Zajac, S. Delimitacja rekreacyjnej funkcji lasów i gospodarki leśnej na terenach zurbanizowanych. Leśne Pr. Badaw. 2011, 72, 83–94. [Google Scholar]
  15. Wang, Z.; Wang, E.; Yu, Y. Translating tourists’ satisfaction data into economic value of the National Forest Parks in China. J. For. Res. 2023, 28, 397–406. [Google Scholar] [CrossRef]
  16. Kim, Y. A guidelines for the media art of forest park using the forms and principles natural art. Int. Soc. Next Gener. Converg. Technol. 2019, 3, 143–149. [Google Scholar] [CrossRef]
  17. Zhang, H.; Yu, J.; Dong, X.; Zhai, X.; Shen, J. Rethinking Cultural Ecosystem Services in Urban Forest Parks: An Analysis of Citizens’ Physical Activities Based on Social Media Data. Forests 2024, 15, 1633. [Google Scholar] [CrossRef]
  18. He, S.; Yu, Y.; Lan, S.; Zheng, Y.; Liu, C. Influence of Perceived Sensory Dimensions on Cultural Ecosystem Benefits of National Forest Parks Based on Public Participation: The Case of Fuzhou National Forest Park. Forests 2024, 15, 1314. [Google Scholar] [CrossRef]
  19. Kang, N.; Wang, E.; Yu, Y. Valuing forest park attributes by giving consideration to the tourist satisfaction. Tour. Econ. 2019, 25, 711–733. [Google Scholar] [CrossRef]
  20. Woodley, S. A scheme for ecological monitoring in national parks and protected areas. Environments 1996, 23, 50–73. [Google Scholar]
  21. Tierney, G.L.; Faber-Langendoen, D.; Mitchell, B.R.; Shriver, W.G.; Gibbs, J.P. Monitoring and evaluating the ecological integrity of forest ecosystems. Front. Ecol. Environ. 2009, 7, 308–316. [Google Scholar] [CrossRef]
  22. Théau, J.; Trottier, S.; Graillon, P. Optimization of an ecological integrity monitoring program for protected areas: Case study for a network of national parks. PLoS ONE 2018, 13, e0202902. [Google Scholar] [CrossRef]
  23. Jianrong, Z.; Zhenbin, Z. Tourists’ perceptual presentation of national forest park—A case study of Wujin mountain national forest park. J. For. Res. 2022, 27, 15–19. [Google Scholar] [CrossRef]
  24. Zeng, Z.X.; Zhang, A.W.; Wang, Q.T. A Research on the Problems and Solutions in the Development of the Forest Parks in China. Appl. Mech. Mater. 2013, 295–298, 2343–2346. [Google Scholar] [CrossRef]
  25. Kozłowska-Adamczak, M.; Jezierska-Thöle, A.; Essing-Jelonkiewicz, P. Application of Remote Sensing for the Evaluation of the Forest Ecosystem Functions and Tourism Services. Sustainability 2025, 17, 2060. [Google Scholar] [CrossRef]
  26. del Castillo, E.M.; García-Martin, A.; Aladrén, L.A.L.; de Luis, M. Evaluation of forest cover change using remote sensing techniques and landscape metrics in Moncayo Natural Park (Spain). Appl. Geogr. 2015, 62, 247–255. [Google Scholar] [CrossRef]
  27. Nagendra, H.; Tucker, C.; Carlson, L.; Southworth, J.; Karmacharya, M.; Karna, B. Monitoring parks through remote sensing: Studies in Nepal and Honduras. Environ. Manag. 2004, 34, 748–760. [Google Scholar] [CrossRef] [PubMed]
  28. Kunakh, O.; Ivanko, I.; Holoborodko, K.; Lisovets, O.; Volkova, A.; Nikolaieva, V.; Zhukov, O. Modeling the spatial variation of urban park ecological properties using remote sensing data. Biosyst. Divers. 2022, 30, 213–225. [Google Scholar] [CrossRef]
  29. Chen, B.; Chen, L.; Huang, B.; Michishita, R.; Xu, B. Dynamic monitoring of the Poyang Lake wetland by integrating Landsat and MODIS observations. ISPRS J. Photogramm. Remote Sens. 2018, 139, 75–87. [Google Scholar] [CrossRef]
  30. Chen, Z.; Chen, J.; Yue, Y.; Lan, Y.; Ling, M.; Li, X.; You, H.; Han, X.; Zhou, G. Tradeoffs among multi-source remote sensing images, spatial resolution, and accuracy for the classification of wetland plant species and surface objects based on the MRS_DeepLabV3+ model. Ecol. Inform. 2024, 81, 102594. [Google Scholar] [CrossRef]
  31. Lingua, F.; Coops, N.C.; Griess, V.C. Assessing forest recreational potential from social media data and remote sensing technologies data. Ecol. Indic. 2023, 149, 110165. [Google Scholar] [CrossRef]
  32. You, S.; Zheng, Q.; Chen, B.; Xu, Z.; Lin, Y.; Gan, M.; Zhu, C.; Deng, J.; Wang, K. Identifying the spatiotemporal dynamics of forest ecotourism values with remotely sensed images and social media data: A perspective of public preferences. J. Clean. Prod. 2022, 341, 130715. [Google Scholar] [CrossRef]
  33. Barros, C.; Moya-Gómez, B.; Gutiérrez, J. Using geotagged photographs and GPS tracks from social networks to analyse visitor behaviour in national parks. Curr. Issues Tour. 2020, 23, 1291–1310. [Google Scholar] [CrossRef]
  34. Walden-Schreiner, C.; Leung, Y.-F.; Tateosian, L. Digital footprints: Incorporating crowdsourced geographic information for protected area management. Appl. Geogr. 2018, 90, 44–54. [Google Scholar] [CrossRef]
  35. Pizam, A.; Milman, A. Predicting satisfaction among first time visitors to a destination by using the expectancy disconfirmation theory. Int. J. Hosp. Manag. 1993, 12, 197–209. [Google Scholar] [CrossRef]
  36. Choi, I.Y.; Moon, H.S.; Kim, J.K. Assessing personalized recommendation services using expectancy disconfirmation theory. Asia Pac. J. Inf. Syst. 2019, 29, 203–216. [Google Scholar] [CrossRef]
  37. Weber, K. The assessment of tourist satisfaction using the expectancy disconfirmation theory: A study of the German travel market in Australia. Pac. Tour. Rev. 1997, 1, 35–45. [Google Scholar]
  38. Ray, R.; Rahman, M.B. Measuring students’ satisfaction towards different tourism destinations in Rajshahi: An Application of Expectancy Disconfirmation Theory. Bangladesh J. Tour. 2016, 1. [Google Scholar]
  39. Tribe, J.; Snaith, T. From SERVQUAL to HOLSAT: Holiday satisfaction in Varadero, Cuba. Tour. Manag. 1998, 19, 25–34. [Google Scholar] [CrossRef]
  40. Puri, G.; Singh, K. The role of service quality and customer satisfaction in tourism industry: A review of SERVQUAL Model. Int. J. Res. Anal. Rev. 2018, 5. [Google Scholar]
  41. Bhattacharya, P.; Mukhopadhyay, A.; Saha, J.; Samanta, B.; Mondal, M.; Bhattacharya, S.; Paul, S. Perception-satisfaction based quality assessment of tourism and hospitality services in the Himalayan region: An application of AHP-SERVQUAL approach on Sandakphu Trail, West Bengal, India. Int. J. Geoheritage Parks 2023, 11, 259–275. [Google Scholar] [CrossRef]
  42. Kouthouris, C.; Alexandris, K. Can service quality predict customer satisfaction and behavioral intentions in the sport tourism industry? An application of the SERVQUAL model in an outdoors setting. J. Sport Tour. 2005, 10, 101–111. [Google Scholar] [CrossRef]
  43. Song, H.J.; Lee, C.-K.; Park, J.A.; Hwang, Y.H.; Reisinger, Y. The influence of tourist experience on perceived value and satisfaction with temple stays: The experience economy theory. J. Travel Tour. Mark. 2015, 32, 401–415. [Google Scholar] [CrossRef]
  44. Lee, S.; Jeong, E.; Qu, K. Exploring theme park visitors’ experience on satisfaction and revisit intention: A utilization of experience economy model. J. Qual. Assur. Hosp. Tour. 2020, 21, 474–497. [Google Scholar] [CrossRef]
  45. Mehmetoglu, M.; Engen, M. Pine and Gilmore’s concept of experience economy and its dimensions: An empirical examination in tourism. J. Qual. Assur. Hosp. Tour. 2011, 12, 237–255. [Google Scholar] [CrossRef]
  46. Mahdzar, M.; Saiful Raznan, A.M.; Ahmad Jasmin, N.; Abdul Aziz, N.A. Exploring relationships between experience economy and satisfaction of visitors in rural tourism destination. J. Int. Bus. Econ. Entrep. (JIBE) 2020, 5, 69–75. [Google Scholar]
  47. Babolian Hendijani, R. Effect of food experience on tourist satisfaction: The case of Indonesia. Int. J. Cult. Tour. Hosp. Res. 2016, 10, 272–282. [Google Scholar] [CrossRef]
  48. Anaya-Aguilar, R.; Gemar, G.; Anaya-Aguilar, C. Validation of a satisfaction questionnaire on spa tourism. Int. J. Environ. Res. Public Health 2021, 18, 4507. [Google Scholar] [CrossRef]
  49. Suzuki, K.; Rin, U.; Maeda, Y.; Takeda, H. Forest cover classification using geospatial multimodal data. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2018, 42, 1091–1096. [Google Scholar] [CrossRef]
  50. Lamahewage, S.H.G.; Witharana, C.; Riemann, R.; Fahey, R.; Worthley, T. Aboveground biomass estimation using multimodal remote sensing observations and machine learning in mixed temperate forest. Sci. Rep. 2025, 15, 31120. [Google Scholar] [CrossRef] [PubMed]
  51. Zheng, P.; Fang, P.; Wang, L.; Ou, G.; Xu, W.; Dai, F.; Dai, Q. Synergism of multi-modal data for mapping tree species distribution—A case study from a mountainous forest in southwest china. Remote Sens. 2023, 15, 979. [Google Scholar] [CrossRef]
  52. Ling, Q.; Chen, Y.; Feng, Z.; Pei, H.; Wang, C.; Yin, Z.; Qiu, Z. Monitoring Canopy Height in the Hainan Tropical Rainforest Using Machine Learning and Multi-Modal Data Fusion. Remote Sens. 2025, 17, 966. [Google Scholar] [CrossRef]
  53. Yang, Y.; Wang, Z.; Lin, G. Performance assessment indicators for comparing recreational services of urban parks. Int. J. Environ. Res. Public Health 2021, 18, 3337. [Google Scholar] [CrossRef]
  54. Jalilian, M.A.; Danehkar, A.; Fami, H.S.A. Determination of indicators and standards for tourism impacts in protected Karaj River, Iran. Tour. Manag. 2012, 33, 61–63. [Google Scholar] [CrossRef]
  55. Lin, S.; Shi, W.; Dong, L. Research on travel decision-making based on text analysis of travel notes—Take Ctrip as a example. In Proceedings of the 2016 IEEE Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC), Xi’an, China, 3–5 October 2016; pp. 859–862. [Google Scholar]
  56. Cohen, J.B.; Fishbein, M.; Ahtola, O.T. The nature and uses of expectancy-value models in consumer attitude research. J. Mark. Res. 1972, 9, 456–460. [Google Scholar] [CrossRef]
  57. Wu, J.; Yang, T. Service attributes for sustainable rural tourism from online comments: Tourist satisfaction perspective. J. Destin. Mark. Manag. 2023, 30, 100822. [Google Scholar] [CrossRef]
  58. Song, S.; Kawamura, H.; Uchida, J.; Saito, H. Determining tourist satisfaction from travel reviews. Inf. Technol. Tour. 2019, 21, 337–367. [Google Scholar] [CrossRef]
  59. Atsri, K.H.; Abotsi, K.E.; Kokou, K.; Dendi, D.; Segniagbeto, G.H.; Fa, J.E.; Luiselli, L. Ecological challenges for the buffer zone management of a West African National Park. J. Environ. Plan. Manag. 2020, 63, 689–709. [Google Scholar] [CrossRef]
  60. Guan, C.; Song, J.; Keith, M.; Akiyama, Y.; Shibasaki, R.; Sato, T. Delineating urban park catchment areas using mobile phone data: A case study of Tokyo. Comput. Environ. Urban Syst. 2020, 81, 101474. [Google Scholar] [CrossRef]
  61. McHugh, M.L. Interrater reliability: The kappa statistic. Biochem. Medica 2012, 22, 276–282. [Google Scholar] [CrossRef]
  62. Holle, H.; Rein, R. The modified Cohen’s kappa: Calculating interrater agreement for segmentation and annotation. In Understanding Body Movements: A Guide to Empirical Research on Nonverbal Behavior: With an Introduction to the NEUROGES Coding System; Peter Lang GmbH: Berlin, Germany, 2013; pp. 261–277. [Google Scholar]
  63. Chen, T. XGBoost: A Scalable Tree Boosting System. arXiv 2016, arXiv:1603.02754. [Google Scholar] [CrossRef]
  64. Lundberg, S.M.; Lee, S.-I. A unified approach to interpreting model predictions. In Proceedings of the NIPS’17: Proceedings of the 31st International Conference on Neural Information Processing Systems, Long Beach, CA, USA, 4–9 December 2017. [Google Scholar]
  65. Blazeska, D.; Strezovski, Z.; Klimoska, A.M. The influence of tourist infrastructure on the tourist satisfaction in Ohrid. UTMS J. Econ. 2018, 9, 85–93. [Google Scholar]
  66. Khaliji, M.A.; Ghalehteimouri, K.J. Assessing tourist infrastructure using a multi-criteria decision-making model: A case study of Ardabil Province’s impact on regional tourism development. Res. Sq. 2024. [Google Scholar]
  67. Priyana, E.B.; Prihartanto, E. The Role of Infrastructure in Realizing Cultural Tourism in North Kalimantan: A Literature Review. IOP Conf. Ser. Earth Environ. Sci. 2024, 1431, 012001. [Google Scholar] [CrossRef]
  68. Kanwal, S.; Rasheed, M.I.; Pitafi, A.H.; Pitafi, A.; Ren, M. Road and transport infrastructure development and community support for tourism: The role of perceived benefits, and community satisfaction. Tour. Manag. 2020, 77, 104014. [Google Scholar] [CrossRef]
  69. Arabov, N.; Nasimov, D.; Janzakov, B.; Khomitov, K.; Utemuratova, G.; Abduraimov, D.; Ismailov, B. Shaping the future of Uzbekistan’s tourism: An in-depth analysis of infrastructure influence and strategic planning. J. East. Eur. Cent. Asian Res. (JEECAR) 2024, 11, 53–65. [Google Scholar] [CrossRef]
  70. Sharma, M.; Mohapatra, G.; Giri, A.K. Assessing the role of ICT, governance, and infrastructure on inbound tourism demand in India. J. Econ. Adm. Sci. 2025, 41, 320–335. [Google Scholar] [CrossRef]
  71. Fanelli, G.; Tescarollo, P.; Testi, A. Ecological indicators applied to urban and suburban floras. Ecol. Indic. 2006, 6, 444–457. [Google Scholar] [CrossRef]
  72. Zhang, T.; Yang, R.; Yang, Y.; Li, L.; Chen, L. Assessing the urban eco-environmental quality by the remote-sensing ecological index: Application to Tianjin, North China. ISPRS Int. J. Geo-Inf. 2021, 10, 475. [Google Scholar] [CrossRef]
  73. Rebele, F. Urban ecology and special features of urban ecosystems. Glob. Ecol. Biogeogr. Lett. 1994, 4, 173–187. [Google Scholar] [CrossRef]
  74. Buckley, R. Ecological indicators of tourist impacts in parks. J. Ecotourism 2003, 2, 54–66. [Google Scholar] [CrossRef]
  75. Deng, J.; Qiang, S.; Walker, G.J.; Zhang, Y. Assessment on and perception of visitors’ environmental impacts of nature tourism: A case study of Zhangjiajie National Forest Park, China. J. Sustain. Tour. 2003, 11, 529–548. [Google Scholar] [CrossRef]
  76. Qin, G.; Cheng, B. Analysis on the impact of Forest Park facilities on the performance of Forest Park tourism: An empirical study of Forest parks in China. Tour. Plan. Dev. 2021, 18, 457–478. [Google Scholar] [CrossRef]
  77. Yu, M.; Liu, Y. Landscape Ecological Integrity Assessment to Improve Protected Area Management of Forest Ecosystem. Ecologies 2025, 6, 38. [Google Scholar] [CrossRef]
  78. Brovina, F.; Sallaku, D. Sustainable development of forest parks for active recreation: A balance between nature conservation and physical education. Ukr. J. For. Wood Sci. 2024, 15, 165–179. [Google Scholar] [CrossRef]
  79. Huang, Z.; Cao, J.; Peng, Y.; Ma, K.; Cui, G. Quantitative Evaluation of the Integrity of Natural Ecosystems and Anthropogenic Impacts in Shennongjia National Park, China. Forests 2023, 14, 987. [Google Scholar] [CrossRef]
  80. Reining, C.E. Does Perceived Ecological Integrity Affect Restorative Health Outcomes? An Examination of Visitor Experiences in Diverse Environments in an Ontario Protected Area. Master’s Thesis, Wilfrid Laurier University, Waterloo, ON, Canada, 2019. [Google Scholar]
  81. Chakraborty, A.; Messias, J.; Benevenuto, F.; Ghosh, S.; Ganguly, N.; Gummadi, K. Who makes trends? understanding demographic biases in crowdsourced recommendations. In Proceedings of the International AAAI Conference on Web and Social Media, Montreal, QC, Canada, 15–18 May 2017; Volume 11, pp. 22–31. [Google Scholar]
  82. Cesare, N.; Grant, C.; Nsoesie, E.O. Understanding demographic bias and representation in social media health data. In Proceedings of the Companion Publication of the 10th ACM Conference on Web Science, Amsterdam, The Netherlands, 30 June–3 July 2019; pp. 7–9. [Google Scholar]
  83. Henderson, K.E.; Welsh, E.T. Potential bias when using social media for selection: Differential effects of candidate demographic characteristics, race match, perceived similarity, and profile detail. Int. J. Sel. Assess. 2024, 32, 149–167. [Google Scholar] [CrossRef]
Sustainability 18 01936 g001
Figure 1. Distribution map of forest parks in the Yangtze River Delta.
Figure 1. Distribution map of forest parks in the Yangtze River Delta.
Sustainability 18 01936 g001
Sustainability 18 01936 g002
Figure 2. Data processing flowchart.
Figure 2. Data processing flowchart.
Sustainability 18 01936 g002
Sustainability 18 01936 g003
Figure 3. Descriptive statistical and spatial analysis maps: (a) Proportion of review popularity (b) Proportion of tourist sentiment.
Figure 3. Descriptive statistical and spatial analysis maps: (a) Proportion of review popularity (b) Proportion of tourist sentiment.
Sustainability 18 01936 g003
Sustainability 18 01936 g004
Figure 4. Numerical and spatial distribution of recreation service indicators: (a) shopping facilities (b) scientific and educational facilities (c) scientific and educational diversity (d) public security agencies (e) road network density (f) parking lots (g) medical institutions (h) landscape ancillary facilities (i) infrastructure update time (j) total sentiment (k) fire departments (l) transportation facilities (m) total review quantity (n) vegetation diversity (o) vegetation density (p) accommodation facilities (q) bus stops.
Figure 4. Numerical and spatial distribution of recreation service indicators: (a) shopping facilities (b) scientific and educational facilities (c) scientific and educational diversity (d) public security agencies (e) road network density (f) parking lots (g) medical institutions (h) landscape ancillary facilities (i) infrastructure update time (j) total sentiment (k) fire departments (l) transportation facilities (m) total review quantity (n) vegetation diversity (o) vegetation density (p) accommodation facilities (q) bus stops.
Sustainability 18 01936 g004
Sustainability 18 01936 g005
Figure 5. Distribution of Local Moran’s I for the (a) comment dimension and (b) sentiment dimension.
Figure 5. Distribution of Local Moran’s I for the (a) comment dimension and (b) sentiment dimension.
Sustainability 18 01936 g005
Sustainability 18 01936 g006
Figure 6. Correlation analysis plot of recreational service indicators and visitor satisfaction.
Figure 6. Correlation analysis plot of recreational service indicators and visitor satisfaction.
Sustainability 18 01936 g006
Sustainability 18 01936 g007
Figure 7. The contribution of each indicator (a) popularity of comments and (b) sentiment in the comments.
Figure 7. The contribution of each indicator (a) popularity of comments and (b) sentiment in the comments.
Sustainability 18 01936 g007
Sustainability 18 01936 g008
Figure 8. Analysis of the Marginal Effects of Recreation Service Indicators on Comment Popularity.
Figure 8. Analysis of the Marginal Effects of Recreation Service Indicators on Comment Popularity.
Sustainability 18 01936 g008
Sustainability 18 01936 g009
Figure 9. Analysis of the Marginal Effects of Recreation Service Indicators on Tourist Sentiment.
Figure 9. Analysis of the Marginal Effects of Recreation Service Indicators on Tourist Sentiment.
Sustainability 18 01936 g009
Table 1. Classification, Definition, and Data Sources of Forest Park Recreation Service and Tourist Satisfaction Indicators.
Table 1. Classification, Definition, and Data Sources of Forest Park Recreation Service and Tourist Satisfaction Indicators.
Items (Abbreviation)DescriptionCalculation MethodSource
education, culture and experiential services indicators
scientific and educational diversity
(SED)		Number of categories of science and environmental education resources within the forest parks (count). S E D = − 1 ∑ i = 1 s p i ln p i
p i = n i N
S = number of education resource categories; n i = number of resources in category i; N = ∑ i = 1 s n i 		Baidu Maps POI data (2024)
scientific and educational facilities
(SEF)		Number of facilities supporting ecological education and environmental interpretation (e.g., nature education centers, interpretive signage, and research monitoring stations) within the forest park buffer zone (count). X m = ∑ i = 1 M n i
m = Total number of facilities in category m
M = Number of sub-types in this facility category
n i = Number of facilities in sub-type
tourism services and supporting facilities indicators
accommodation facilities
(AF)		Number of accommodation facilities (e.g., hotels, homestays, campsites) within the forest park buffer zone (count).
landscape ancillary facilities
(LAF)		Number of facilities enhancing landscape presentation and visitor viewing experience within the forest park buffer zone (count).
shopping facilities
(SF)		Number of retail, food, and cultural–creative facilities within the forest park buffer zone (count).
infrastructure and accessibility indicators
transportation facilities
(TF)		Number of facilities supporting visitor access and internal circulation within the forest park buffer zone (count).
road network density
(RND)		Density of the road network within the buffer zone (km/km2).RND = L A
L = total road length within the buffer zone (km)
A = buffer zone area (km2) 		Open-Street Map (OSM)
bus stops
(BS)		Number of bus stops serving the forest park within the buffer zone. (count) X m = ∑ i = 1 M n i
m = Total number of facilities in category m
M = Number of sub-types in this facility category
n i = Number of facilities in sub-type		Baidu Maps POI data (2024)
parking lots
(PL)		Number of parking facilities for motorized and non-motorized vehicles within the buffer zone (count).
safety and operational assurance indicators
public security agencies
(PSA)		Number of public security facilities (e.g., police stations, public security bureaus, police substations, and joint public security patrol posts) within the forest park buffer zone (count).
medical institutions
(MI)		Number of medical and emergency service facilities within the forest park buffer zone (count).
fire departments
(FD)		Number of firefighting facilities and fire stations within the forest park buffer zone (count).
infrastructure update time
(IUT)		Year of the most recent construction or renovation of internal park infrastructure (year).IUT =   Y e a r c u r r e n t − Y e a r c o n s t r u c t i o n
Y e a r c o n s t r u c t i o n = construction or renovation dates of internal park facilities.		The government websites (2024)
ecological environmental quality indicators
vegetation diversity
(VDEN)		Vegetation species diversity per unit area within the forest park (species/ha).VDEN = S A
S = number of vegetation species
A = park area (ha)		Field survey data (2024)
vegetation density
(VDIV)		The ratio of the total vegetation area to the total buffer area (%).VDIV = A v e g A b u f f e r
A v e g = the total area covered by vegetation within the buffer zone. A b u f f e r = denotes the total area of the buffer zone.		Remote sensing data (2024 annual mean NDVI data)
tourist satisfaction indicators
total review quantity
(TRQ)		The number of comments extracted from social media using text mining in forests park(count).TRQ = ∑ C o u n t c o m m e n t s
C o u n t c o m m e n t s = number of comments extracted from social media using text mining.		Web-scraped data from the social media platform Ctrip
tourism service review quantity
(TSRQ)		Number of social media reviews related to tourism services (count).TSRQ = ∑ C o u n t t o u r i s m _ r e v i e w s
C o u n t t o u r i s m _ r e v i e w s = number of social media reviews related to tourism services.
recreation service review ratio
(RSRR)		The proportion of recreation service -related reviews to total reviews (%). R j = N j T R Q × 100 %
Rj = The proportion of comments on a certain type of service
Nj = Number of reviews related to class j service
TRQ = Total number of visitor reviews extracted from social media platforms.
science and education service review ratio
(SERR)		The proportion of science and education service-related reviews to total reviews (%).
public facility review ratio
(PFRR)		The proportion of public facility-related reviews to total reviews (%).
transportation review ratio
(TRR)		The proportion of transportation review to total reviews (%).
safety service review ratio
(SRR)		The proportion of safety service review to total reviews (%).
landscape review ratio
(LRR)		The proportion of landscape reviews to total reviews (%).
total sentiment
(TE)		Aggregated sentiment score derived from sentiment analysis of visitor reviews. T E = ∑ i = 1 T R Q S c o r e i
S c o r e i   = sentiment score of the i-th visitor comment derived from sentiment analysis.
negative sentiment
(NEG)		Number of reviews expressing negative sentiment (count). S k = C o u n t k
S k = The total quantity of the k type of sentiment
C o u n t k = The number of social media comments corresponding to the sentiment
neutral sentiment
(NE)		Number of reviews expressing neutral sentiment (count).
positive sentiment
(PE)		Number of reviews expressing positive sentiment (count).
Table 2. Descriptive statistics result of related indicators (N = 67).
Table 2. Descriptive statistics result of related indicators (N = 67).
Indicator[Min, Max]MeanSDMedian
PE[43–3390]848.95945.99375
NE[15–827]231.29218.52137.5
NEG[1–329]75.0683.5532.5
TE[52–3474.5]889.54969.89412
TRQ[91–5611]1350.641396.02634.5
SED[0–10]2.942.353
SEF[0–143]12.0523.44
AF[0–1808]121.3295.8531.5
LAF[1–536]52.3283.3721.5
SF[0–401]42.3976.067.5
TF[0–32]1.614.940
RND[0.157–9.811]1.942.061.1375
BS[0–320]27.6849.8513.5
PL[1–964]85.68155.3415.5
PSA[0–325]34.2158.468.5
MI[0–250]26.1853.373.5
FD[0–52]6.2911.091.5
IUT[1–5]2.581.692
VDEN[1.638–4.154]3.060.793.0305
VDIV[0.309–0.797]0.610.130.616
TSRQ[67–4546]1155.31225.42528.5
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Chen, C.; Zhao, W.; Zhao, B. Mechanisms Underlying the Effects of Recreation Services on Tourist Satisfaction in Forest Parks: A Case Study of the Yangtze River Delta, China. Sustainability 2026, 18, 1936. https://doi.org/10.3390/su18041936

AMA Style

Chen C, Zhao W, Zhao B. Mechanisms Underlying the Effects of Recreation Services on Tourist Satisfaction in Forest Parks: A Case Study of the Yangtze River Delta, China. Sustainability. 2026; 18(4):1936. https://doi.org/10.3390/su18041936

Chicago/Turabian Style

Chen, Caijie, Weilin Zhao, and Bing Zhao. 2026. "Mechanisms Underlying the Effects of Recreation Services on Tourist Satisfaction in Forest Parks: A Case Study of the Yangtze River Delta, China" Sustainability 18, no. 4: 1936. https://doi.org/10.3390/su18041936

APA Style

Chen, C., Zhao, W., & Zhao, B. (2026). Mechanisms Underlying the Effects of Recreation Services on Tourist Satisfaction in Forest Parks: A Case Study of the Yangtze River Delta, China. Sustainability, 18(4), 1936. https://doi.org/10.3390/su18041936

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop