Assessing urban park services for children’s wellbeing using user-generated content (UGC): A case study of Shanghai, China

Abstract

Urban parks play a crucial role in children’s development. With UGC demonstrating their value in capturing granular user experiences, they have become an increasingly important data source in park research. However, their application remains preliminary and challenging when studying children. This study investigates urban park services related to children’s wellbeing through user-generated content from 46 parks in Shanghai. It developed a structured lexicon by integrating 11 established audit tools with UGC, quantifying public perception of services across 9 dimensions (including 44 indicators) of children’s well-being. Through BERT sentiment analysis, regression modeling, and machine learning, this study assessed service performance, identified key services influencing perceptions of children’s wellbeing, and captured their non-linear and interaction patterns. Findings indicate that Education services show substantial variation across parks; Safety, Nature, Comfort, and Play services significantly influence perceptions, with Play dominating in parks with sufficient facilities. Interaction effects between Play-Nature and Play-Safety combinations highlight the importance of integrated design. Additionally, the analysis revealed a conflict between safety management practices and user expectations in Shanghai parks, suggesting a shift from restrictive to supportive safety practices. These findings provide practical insights for park optimization and demonstrate the potential of UGC in child-friendly environment research.

Graphical abstract

Keywords

Children’s wellbeing / Urban parks / User-generated content / Public perception / Shanghai

1 Introduction

Global urbanization has significantly limited children’s access to outdoor spaces (Andal, 2023; Dong and Geng, 2023; Gao et al., 2024), contributing to what Louv (2005) termed "Nature Deficit Disorder". This phenomenon compromises children’s holistic development and wellbeing (Amoly et al., 2014; UNICEF, 2012). Urban parks offer numerous benefits for children by promoting physical health through active play (Christian et al., 2015; Cohen et al., 2020), stimulating cognitive development, enhancing creative thinking (Herrington and Brussoni, 2015), and building social competence through peer interaction (Veitch et al., 2021). However, research shows that the quality of parks directly influences children’s visits and the developmental benefits they derive (Loukaitou-Sideris and Sideris, 2009; Rigolon, 2017), with equitable access to high-quality parks remaining a persistent challenge (Chen et al., 2019). This issue is particularly relevant in China, which has the world’s second-largest child population (298 million as of 2023) (NBSC et al., 2023), and is experiencing rapid urbanization. In response, the Chinese government has issued guidelines to promote child-centered urban spaces (MHURD et al., 2023; NDRC et al., 2021), with particular emphasis on parks and other open spaces. Within this context, it is crucial to understand and assess whether parks have the qualities and features that enhance children’s wellbeing (Pham et al., 2019; Yang et al., 2024).

Traditional park assessment methods for children typically employ questionnaires, interviews, and observations (Aarts et al., 2012; Cohen et al., 2020; Istiani et al., 2023; Saelens et al., 2006), but these approaches require substantial time and human resources, consequently limiting sample sizes, spatial coverage, and data collection periods (Li et al., 2023; Tang et al., 2023). Additionally, these methods can be influenced by the quality of evaluation instruments, such as questionnaires and audit forms, potentially overlooking significant concerns reported by users (Huai et al., 2023; Wan et al., 2021; Wang et al., 2021b). while assessments conducted by researchers ensure standardization and comparability across sites (Heikinheimo et al., 2020; Yang et al., 2024; Zhou et al., 2024), this approach potentially fails to represent long-term usage patterns and comprehensive opinions from diverse users (Park, 2017; Wang et al., 2021a).

Meanwhile, user-generated content(UGC) has emerged as a valuable source in urban studies (Wang et al., 2021a; Zhao et al., 2024; Zhou et al., 2024), due to its spontaneity, volume (Cheng et al., 2021; Komossa et al., 2020), and ability to capture authentic user experiences (Chen et al., 2024; Huai et al., 2023; Li et al., 2023; Rui, 2023; Wilkins et al., 2022; Zhao et al., 2024). When combined with analytical techniques like sentiment analysis, UGC can effectively indicate facility quality with user satisfaction levels, providing a powerful, low-cost approach to under-standing public space perceptions without the resource constraints of traditional methods (Wei et al., 2023). However, despite children being frequently mentioned in park reviews (Zhao et al., 2024), research utilizing UGC to assess park resources for children remains preliminary and challenging (Chen et al., 2024; Yang et al., 2024). This gap limits the potential application of UGC in investigating child-friendly environments, especially considering that children represent vulnerable user groups in terms of park resource provision (Sikorska et al., 2020).

This study aims to utilize UGC to investigate children’s wellbeing in parks, providing evidence for park resource optimization. The specific research questions addressed in this study include:

1) How can UGC be used to assess park services for children’s well-being? This study explores to develop a structured framework with established audit tools and UGC to assess park services relevant to children;

2) What does UGC reveal about these services? This question examines variations in users’ perception to reveal the presence and condition of child-related services across parks;

3) What services influence perception of children’s wellbeing in parks? The study employs multiple regression models and machine learning approaches to identify key services and their interactive effects on users’ perception of children’s well-being in urban park environments.

2 Literature review

2.1 Child-oriented park audit tools and evaluation methods

Several established audit tools are available for identifying and assessing park features related to children’s wellbeing. These tools can be categorized into three types: tools specifically developed for children and adolescents to assess public spaces such as parks (Bird et al., 2015; Crawford et al., 2008; Garau and Annunziata, 2019; Rigolon and Németh, 2018); comprehensive park audit tools that consider children’s needs but are not exclusively child-focused (Kaczynski et al., 2012; Saelens et al., 2006; Veitch et al., 2013), and tools not specifically designed for parks but applicable for assessing children’s play and sports resources within parks (DeBate et al., 2011; Gustat et al., 2019; Jenkins et al., 2015; Woolley and Lowe, 2013).

Despite variations in assessment frameworks and audit items, most tools evaluate park resources including play features (e.g., playgrounds, equipment), sports facilities (e.g., courts, bike paths), natural features (e.g., water bodies, lawns), amenities (e.g., restrooms, seating), and environmental quality aspects (e.g., shade, maintenance). More comprehensive tools additionally assess access (e.g., parking, public transit), safety features (e.g., lighting, surveillance), educational features, and environmental issues such as cigarette litter and noise. However, no single tool comprehensively covers all park features relevant to children. For example, while relatively comprehensive, the Environmental Assessment of Public Recreation Spaces (EAPRS) lacks age-appropriateness assessments for play facilities (Saelens et al., 2006); the Community Park Audit Tool (CPAT) omits certain play facility types (e.g., challenge play equipment, sandbox) (Kaczynski et al., 2014); and the Resilience for Eating and Physical Activity Despite Inequality (READI) park audit tool excludes baby care facilities (Veitch et al., 2013). Therefore, synthesizing elements from across existing tools is essential for comprehensive assessment of how parks support children’s wellbeing.

Methodologically, most audit tools rely on trained assessors conducting on-site observations, which requires significant time and financial resources (Rigolon and Németh, 2018). While the Quality Index of Parks for Youth (QUINPY) adopts public geographic information system (GIS) databases for assessment, this approach faces potential data availability challenges, and many important items cannot be accurately evaluated through remote imagery (Kaczynski et al., 2014).

Most tools primarily evaluate feature presence or quantity, along with usability and condition. Research by Saelens et al. (2006) indicates that assessments of feature presence typically demonstrate higher reliability, whereas more subjective evaluations (e.g., aesthetics, safety) generally show lower inter-rater consistency (Bird et al., 2015). When facing difficult assessment items, auditors may unconsciously favor satisfactory rather than accurate responses (Gustat et al., 2019). Additionally, evaluations often represent temporal "snapshots" rather than continuous assessments (DeBate et al., 2011), resulting in lower reliability for temporally variable features like cleanliness (Saelens et al., 2006). These limitations highlight the need to explore more reliable methods for assessing subjective aspects of park features (Bird et al., 2015).

In addition, these instruments may be biased toward large resources (DeBate et al., 2011). Since evaluation scale affects resource allocation potential, larger parks typically receive higher scores (Meng and Wang, 2022; Woolley and Lowe, 2013).

Most critically, audit scores cannot indicate users’ perceptions of the resources (DeBate et al., 2011). Rigolon and Németh (2018) acknowledge that maintenance levels expressed by city standards might not reflect actual park conditions. Kaczynski et al. (2012) noted that perceptions of park attributes (e.g., safety, quality) are just as or more important than audited features. Future research should explore how evolving technology and direct observation can be combined to gather optimal information about park environmental characteristics (Kaczynski et al., 2012).

2.2 Using UGC to investigate park services

The emergence of social media platforms has created unprecedented opportunities for researchers to access large-scale, UGC about urban spaces. This data revolution has transformed urban studies by providing insights into how people perceive, use, and value public spaces (Dunkel, 2015; Shelton et al., 2015). Social media data offers several distinct advantages over traditional research methods: it captures spontaneous user reactions rather than prompted responses, provides continuous rather than periodic feedback, and encompasses diverse user perspectives at scale (Marti et al., 2019; Tenkanen et al., 2017). In park research specifically, researchers have leveraged various platforms, including review sites like TripAdvisor and Dianping (Zhang and Zhou, 2018), location-based services such as Foursquare (Donahue et al., 2018), photo-sharing platforms like Flickr and Instagram (Sessions et al., 2016), and microblogging sites such as Twitter and Weibo(Guerrero et al., 2016).

Recent studies typically apply statistical and machine learning techniques to extract explicit features from UGC (Zhao et al., 2024). Common approaches include high-frequency word analysis (Liu et al., 2023; Tang et al., 2023) and topic modeling to identify dominant themes (He et al., 2023; Huai and Van de Voorde, 2022; Song et al., 2021). Advances in natural language processing, including BERT-based sentiment classification, have enabled researchers to evaluate satisfaction levels with specific park attributes (Cheng et al., 2021; Huai et al., 2023; Li et al., 2023; Wankhade et al., 2022; Zhou et al., 2024). In parallel, the integration of image recognition and multimodal learning has facilitated "visual-semantic" analyses that enhance perceptual dimension coverage (Chen et al., 2024; Yan et al., 2024; Zhao et al., 2024).

However, despite these methodological innovations, there are significant limitations in utilizing UGC for park service analysis. Current research assumes that environmental features frequently mentioned in reviews are the primary concerns of users (Hausmann et al., 2020; Wan et al., 2021). Given that UGC predominantly reflects experiences with facilities that are available in parks, this approach may lead to what could be termed a "supply masking effect"―where services that are absent or inadequate are less frequently mentioned in reviews. This can lead to misinterpretation where "low mention due to low supply" is incorrectly perceived as "low public interest." As noted by Zhao et al. (2024), the ability to extract perceptions is limited to information explicitly mentioned by visitors, making it difficult to capture insights about dimensions rarely discussed in UGC.

Furthermore, since UGC is passively collected and closely tied to users’ preferences, it cannot be customized to gather specific data as standard surveys can (Yang et al., 2024). UGC processing often functions as a "black box," with interpretations varying between studies. Consequently, even when addressing the same research topic, findings from UGC analyses and traditional surveys often lack comparability (Koblet and Purves, 2020; Wang et al., 2021a). Some studies have addressed this issue by developing lexicons to convert user-generated content into structured, comparable evaluation results (Wang et al., 2021a), with predefined evaluation dimensions ensuring that even infrequently mentioned services are systematically captured. However, UGC is produced by self-selected users, often with varying motivations and demographic profiles, which can introduce representational bias (Pokhriyal et al., 2023).

In this context, this study employs UGC not to directly substitute children’s or parents’ perceptions, but rather to use publicly available comments to infer whether child-related services are widely present in parks. Moreover, the emotional tone in user reviews can offer signals about the physical state or functional quality of such facilities. Although the general public may not fully represent target user groups (e.g., children and parents), they are nevertheless legitimate users of public spaces. Therefore, their perception toward child-related amenities can still serve as useful indirect indicators.

3 Materials and methods

3.1 Study design

This study employs a mixed-methods approach to investigate park services for children’s well-being through UGC analysis (Fig. 1). First, a structured lexicon is developed by integrating established audit tools to transform unstructured UGC into structured, quantifiable data, enabling systematic assessment of park services for children from multiple dimensions. Second, the study analyzes the distribution characteristics of mention frequency across parks to capture the provision differences of park services. The performance of these services is assessed using BERT-based sentiment analysis. Keyword-in-Context (KWIC) and co-occurrence network analysis are used to explore underlying causes of dissatisfaction, while Importance-Performance Analysis identifies parks requiring priority improvements in specific service dimensions. Then, multiple regression models are used to identify services significantly influencing park perception of children’s well-being, along with a machine learning model to capture the relational effects.

3.2 Data sources overview

3.2.1 Data platform and collection method

UGC in this study refers to unstructured textual reviews posted by users on review platforms about their park experiences. This study utilizes UGC data from Dianping, a popular Chinese review platform where users can share reviews of parks and other facilities (Huai et al., 2023; Liu et al., 2023; Wang et al., 2021a; Zhao et al., 2024). Compared to other platform such as Ctrip, whose users are primarily tourists, users of Dianping are mainly local residents (Huai et al., 2023), providing insights into authentic local perspectives.

User reviews were scraped from Dianping using Python. Each review includes the posting date, textual content, and customer experience rating (1–5 stars). For parks that underwent major renovations during this period, only reviews posted at least one month after renovation completion were included to mitigate potential bias from short-term novelty effects.

3.2.2 Park selection criteria

This study focuses on freely accessible urban parks, excluding fee-based attractions (zoos, botanical gardens, scenic parks) and specialized facilities (forest parks, martyr cemeteries) to ensure the sample captures daily usage of children and their families in urban environments. Parks with at least 100 reviews were selected to ensure sufficient data for analysis. The selected parks represent diverse characteristics, including: (1) various types and scales (city, community, and pocket parks); (2) different sizes; (3) varying star ratings; (4) different administrative districts; and (5) varied location zones.

3.3 Structured lexicon development process

3.3.1 Lexicon framework construction

This section describes the development of a structured lexicon that enables the systematic mapping of unstructured UGC into predefined service dimensions. The lexicon serves as the foundation for transforming diverse user expressions into standardized, analyzable categories.

This study synthesized 11 audit tools (C-POST, PARK, QUINPY, OCUS, CPAT, READI park audit tool, EAPRS, PARA, Woolley and Lowe’s play space assessment tool, PSQAT, PSAT) selected through a systematic literature review to extract park features related to children’s wellbeing (Meng and Wang, 2022).

All park features related to children were extracted from these tools, excluding those inapplicable to the Chinese context (e.g., barbecue facilities, swimming pools) and ambiguous elements that could cause semantic analysis errors (e.g., pet dogs; negative reviews from dog owners regarding pet restrictions are often viewed positively by parents as safety measures for children, with such opposing viewpoints leading to contradictory analytical outcomes).

The original audit tools comprised between 3 and 16 dimensions. Through thematic synthesis of dimension names and definitions from the original tool (e.g., merging "Play set or structure features" from EAPRS, "Play value" from PSQAT, and "Structured play diversity" from QUINPY into a single "Play" dimension), and consultation with child-friendly environment experts, nine core dimensions were established.

Additionally, terminology from several authoritative documents was incorporated, including the No Power Type of Amusement Devices―Terminology (SAC, 2012)), Amusement Devices Terminology (SAC, 2017), Safety for Outdoor Body-Building Equipment―General Requirements (SAC, 2011), and Sport Information Classification and Code―Part 5: Sport Venue Code (SAC, 2024). These sources enriched the seed words related to children’s play facilities and fitness equipment.

3.3.2 Vocabulary expansion

The vocabulary expansion process followed a systematic approach beginning with data cleaning, which included: (1) removing duplicate reviews; (2) eliminating irrelevant content such as URLs, emojis, and advertisements; and (3) excluding reviews containing fewer than four Chinese words, following established methodologies in Chinese text mining research (Shi et al., 2022; Zhao et al., 2020). These ultra-short reviews lack sufficient contextual information and contain minimal semantic value, potentially reducing the precision of sentiment analysis and feature identification. This threshold excluded approximately 0.01% of the original dataset, with analysis showing no systematic exclusion of specific demographic groups.

The preprocessed reviews were segmented using the Chinese word segmentation tool (Sun, 2020), followed by stop word removal. A custom dictionary containing park and attraction names was constructed for additional filtering. The Term Frequency-Inverse Document Frequency (TF-IDF) method was employed to generate characteristic words, mitigating interference from non-distinctive high-frequency words (e.g., "park," "children"). As users could express identical features using different synonyms, the pre-trained Chinese Word2Vec model from Tencent AI Lab was utilized for further vocabulary expansion. This model includes 8 million Chinese words trained on extensive internet textual data, words with cosine similarity values exceeding 0.7 relative to seed words were identified. All potential additions underwent screening before incorporation into the final dictionary.

3.3.3 Lexicon validation and optimization

To validate the dictionary’s effectiveness, a random sample of 1000 short reviews was manually annotated by two researchers independently. The manually annotated results were compared with the automatic identification results, and the accuracy, recall, and F1 score were calculated for the overall results and each dimension. To ensure the reliability of manual annotation, the Cohen’s Kappa coefficient between annotators was computed. For reviews with inconsistent annotations, experts reviewed and finalized the labels. Vocabulary for dimensions with accuracy or recall rates below 0.70 was adjusted and supplemented to produce the final dictionary.

3.4 Perception frequency distribution analysis

This section describes how unstructured review text is transformed into quantifiable perception frequency data through systematic word-matching procedures.

The word-matching approach was employed to identify the frequency and count of service mentions across parks. Considering that each user review may contain evaluations across multiple dimensions, the text was first segmented into sentences based on sentence terminators such as periods, question marks, exclamation points, and ellipses, generating a total of 109,532 short reviews. Reviews matching dictionary words were included in corresponding evaluation items; those without matches were treated as invalid comments (Wang et al., 2021a). Each review could relate to multiple evaluation items, but each item was recorded only once per match. Perception frequency represents the proportion of reviews mentioning a specific service dimension and was calculated using Eq. (1):

\begin{array}{}\text{(1)}& P_{ni}=\frac{C_{ni}}{R_n}\mathrm{.}\end{array}

Where P_ni represents the perception frequency of indicator i in park n, C_ni represents the matching count of indicator i in park n, and R_n represents the total number of valid comments for park n.

Distribution characteristics of perception frequency were utilized to capture the provision differences of park service.

1) Distribution Characteristic Analysis. The Shapiro-Wilk test determined whether perception frequency of each service followed normal distribution. Descriptive statistics (mean, standard deviation, coefficient of variation, skewness) were calculated to assess central tendency, dispersion, and distribution shape of perception frequency.

2) Significance Testing of Differences. Based on normality test results, pairwise comparisons of service perception frequency were conducted using paired-sample t-tests for normally distributed services and Wilcoxon signed-rank tests for non-normally distributed services. This analytical approach enables validation of the classification of park features related to children’s well-being through statistical comparison of user perceptions across different service dimensions.

3.5 Sentiment analysis and service performance assessment

Sentiment analysis is a computational technique that automatically identifies and quantifies emotional attitudes expressed in text. In this study, sentiment analysis is applied to evaluate user satisfaction with the identified service mentions to assess service performance.

Specifically, Bidirectional Encoder Representations from Transformers (BERT) was employed for sentiment analysis. BERT, a pre-trained semantic representation model developed by Google AI in 2018 (Devlin et al., 2019), quantifies reviewer satisfaction with demonstrated superior performance in sentiment analysis than other models (Zhou et al., 2024). BERT-based sentiment analysis scored the segmented short reviews, and average sentiment scores for each service in each park were calculated.

To explore reasons for dissatisfaction, Keyword-in-Context (KWIC) analysis was combined with co-occurrence network analysis. Short reviews with sentiment scores below 1.25 (on a scale of 1–2) were selected, and the top 10 most frequently occurring words within a fixed window of ±5 words around the dictionary terms were extracted. A word co-occurrence network of these extracted terms was then constructed to reveal the primary reasons for dissatisfaction.

To pinpoint parks that require priority improvement, importance-performance analysis (IPA) recognized parks with high perception but poor service performance. IPA is a comprehensive evaluation method used to analyze service quality and has recently been widely applied in park research (Chen et al., 2024; Martilla and James, 1977). IPA analyzes service quality based on importance and satisfaction matrices divided into four quadrants (Fig. 2): Quadrant I (high importance, high satisfaction) represents areas to maintain; Quadrant II (low importance, high satisfaction) represents areas requiring little improvement; Quadrant III (low importance, low satisfaction) represents low-priority areas; and Quadrant IV (high importance, low satisfaction) identifies areas needing major improvement (Bi et al., 2019; Matzler et al., 2004). In this study, perception frequency served as the importance score and sentiment scores as satisfaction scores to identify parks in Quadrant IV for each service.

3.6 Service impact analysis on park perception of children’s well-being

To identify services significantly affecting park perception of children’s well-being, a robust regression model established the relationship between review count and perception count, generating a “predicted perception count” line reflecting the average perception level of children’s well-being of parks based on review count. Robust regression (RLM) reduced the impact of extreme values of review count on fitting results. Relative residuals between actual and predicted perception values were calculated to determine parks’ support level for children’s well-being. Higher relative residuals indicate that the public can perceive more features of children’s well-being in these parks.

Relative residuals were used as the dependent variable y, with the perception frequencies of services as independent variables. Ordinary Least Squares (OLS), Least Absolute Shrinkage and Selection Operator (LASSO), and Ridge models were used to identify services relevant to park perception and compare their relative contributions. The LASSO regression model was constructed with cross-validation (LassoCV), selecting the optimal regularization parameter through 5-fold cross-validation. The Ridge model was implemented with cross-validation (RidgeCV) to select the optimal regularization parameter alpha in logarithmic space, effectively mitigating multicollinearity issues and obtaining more stable coefficient estimates.

To explore potential non-linear relationships between services and the dependent variable, a Support Vector Regression (SVR) model was employed. The SVR model is a non-parametric machine learning approach that effectively captures non-linear relationships, particularly with complex patterns and small datasets. Additionally, univariate and bivariate partial dependence plots (PDP) explored the impact of various services on the outcome.

3.7 Selection of the case study

Shanghai represents a paradigmatic case of Chinese urbanization, characterized by one of the country’s lowest total fertility rates (0.6) and a proportion of residents aged 0–14 years (9.8%) substantially below the national average (16.9%). In response to these challenges, the Government issued the IPCCFCS (GOSMPG, 2022) and the TYAPCFCC (SMDRC, 2023). These initiatives emphasize prioritizing children’s recreational needs in parks and green areas and establishing 50 child-friendly parks. Furthermore, Shanghai’s active social media landscape provides rich user-generated data for this study.

From an initial pool of 488 urban parks sourced from SLCAAB (2024), the final sample comprised 46 parks (Fig. 3) representing diverse characteristics as outlined in the park selection criteria. The spatial distribution of these parks across Shanghai’s urban area allows for comprehensive analysis of park services across different urban contexts. These 46 parks contained a total of 36879 reviews, ranging from 103 to 7565 reviews per park, with an average of 802 reviews per park. This distribution ensured sufficient data volume for robust analysis across all parks while capturing diverse usage patterns across the city. Parks with higher visitor numbers generally had more reviews, but the statistical techniques employed in this study (e.g., robust regression models, relative residuals) control for this variation in sample size.

4 Results

4.1 Lexicon structure and performance

The lexicon covered a total of 2057 words in its vocabulary, including nine dimensions of children’s wellbeing in parks, which were divided into 44 indicators (Table 1). Of these, 41 indicators were extracted from audit tools and 3 indicators were incorporated from user reviews, including power amusement rides, book and reading facilities, and mosquitoes. Fig. 4 illustrates the relationship between the original audit tools and lexicon structure in this study.

The lexicon showed reliable performance metrics (Table 2), with an overall accuracy of 82.4%, recall of 84.8%, and F1 score of 83.5%. Dimension-specific F1 scores ranged from 77.5% (Amenities) to 92.3% (Hygiene). The Cohen’s Kappa coefficient (0.935 overall, ranging from 0.889 to 0.958 across dimensions) suggests strong inter-annotator reliability.

4.2 Distribution characteristics of service perception

Statistical analysis revealed distinct distribution patterns among park services. Shapiro-Wilk tests showed Access (p = 0.1490), Nature (p = 0.1909), and Hygiene (p = 0.1285) followed normal distributions, while other services exhibited right-skewed distributions. Access had the highest mean perception frequency (0.434), followed by Nature (0.262) and Play (0.241), with Safety (0.038) and Education (0.049) being the least perceived (Fig. 5).

Standard deviations showed that Nature (0.660) and Play services (0.577) had the largest perception variations across parks, while Safety (0.023) and Hygiene (0.025) showed the smallest differences. The coefficient of variation suggested that Education services (1.349) had the most unevenly distributed perception across parks, while Access services (0.200) were most uniformly perceived.

Amenities (2.434) and Education (2.298) had the highest skewness values. The box plot showed that Education had discrete high values (Fig. 6), indicating that while most parks received little discussion about educational services, a few parks generated significantly more mentions.

Significance tests revealed distinct perceptions between most service pairs, with 33 of 36 possible comparisons showing significant differences (p < 0.05), except 3 pairs: Play and Nature (p = 0.416), Play and Comfort (p = 0.084), and Education and Safety (p = 0.861). These results support the discriminant validity of the nine-dimension framework, indicating that users perceive these services as distinct aspects of their park experience.

4.3 Sentiment scores and IPA results

Results showed that public satisfaction with services is highest for Education, followed by Nature and Comfort. Safety has the lowest satisfaction, with Access service ranking slightly better. Hygiene and Safety services showed significant variability in sentiment scores across parks (Fig. 7). Figure 8 further illustrated the performance of indicators across services. Within Sports service, wheeled play-friendly features (1.586) demonstrated the lowest satisfaction. Comfort service showed deficiencies in stroller accessibility (1.511). Hygiene dissatisfaction stemmed primarily from environmental issues including mosquitoes (1.515), pollution (1.382), smoking/cigarette (1.327), and noise (1.393). All indicators of Safety consistently received low satisfaction scores, while Parking (1.525) emerged as the primary deficiency within Access services.

Keyword-in-context (KWIC) analysis and co-occurrence network results (Fig. 9) revealed the underlying causes of user dissatisfaction. Findings indicate that low satisfaction with Wheeled Play-Friendly facilities primarily stems from management policies―keywords such as "Enter," "Not allowed," "Children," "Bike," and "Scooter" suggest restrictions on scooters entering parks, limiting important outdoor activities for children. In Safety Management, high-frequency terms including "Say," "Attitude," "No," and "Manner" reflect communication issues with management personnel. Similarly, Stroller Accessibility concerns also highlighted "management" problems, indicating conflicts between administrative policies and user needs. Additionally, results identified safety hazards in facilities, insufficient parking spaces with associated fee issues, and inadequate signage design.

The IPA results identified parks needing priority improvements across service dimensions (Fig. 10). Key deficiencies include play services in Park 42; sports services in Parks 23, 21, and 19; amenities in Parks 23 and 26; comfort services in Parks 19, 21, and 7; hygiene services in Park 18; safety in Parks 7 and 21; and access services in Parks 18 and 21. Several parks exhibited deficiencies in multiple areas, with Parks 18 and 21 each appearing in five service dimensions, while Parks 7, 19, 33, 42, 23, 13, and 24 each appeared in four. This pattern suggests a need for comprehensive service improvements in these parks to better support children’s wellbeing.

4.4 Service impact on park perception of children’s well-being

4.4.1 Regression analysis of service impact

Regression analysis consistently revealed services significantly associated with relative residuals across all three models (Table 3). These models showed good fit (R² = 0.681 OLS, 0.677 LASSO, 0.669 Ridge) with no multicollinearity among services (VIF < 3.581).

OLS analysis identified four services with significant effects: Safety (9.609, p < 0.001), Comfort (1.871, p < 0.05), Nature (1.027, p < 0.05), and Play (0.989, p < 0.05). LASSO and Ridge models confirmed these directional relationships. Safety demonstrated the strongest impact despite its lowest perception frequency. Among the three normally distributed services, only Nature showed significant influence, despite Access service having the highest average perception frequency. This suggests that while access is widely recognized (Curtis et al., 2022), it may function as a baseline requirement rather than a differentiating factor.

Notably, Access services, though most frequently perceived, had no significant impact, while Safety services, mentioned least frequently, exerted the strongest influence.

4.4.2 Partial dependence analysis of service interactions

Univariate Partial Dependence Plots (PDPs) from the SVR model showed how each service perception influenced the dependent variable while controlling for other features (Fig. 11). Play showed minimal impact at lower values but a notable upward trend starting around 0.3. Nature demonstrated substantial influence of approximately 0.4, after which the curve plateaued, suggesting limited additional benefits from further increases. Safety perception displayed a consistent upward trend beyond 0.04, while Comfort, despite its overall positive influence, showed diminishing marginal returns as values increased.

Bivariate PDPs revealed complex interaction effects between service pairs (Fig. 12). The Play-Nature interaction showed a notable pattern: at low Play levels, Nature had relatively pronounced effects; however, as Play values increased, Nature’s marginal effect diminished with Play increasingly dominating perception. Despite this transition, their combined high levels produced apparent synergistic effects. The Play-Safety PDP exhibited consistent synergistic effects, with both variables mutually enhancing each other’s impact on perception. In contrast, the Play-Comfort PDP revealed a localized depression in the response surface (forming a “valley-like” pattern) when Play reached a moderate level but Comfort remained low, indicating that this specific combination produces lower perception values than other combinations. However, when Play values exceeded 0.4, Play emerged as the dominant factor regardless of Comfort levels. These interaction patterns revealed that play emerged as particularly influential at higher levels across all service pairs, and certain service combinations exhibited synergistic effects beyond their individual contributions.

5 Discussion

5.1 Contributions of the structured textual lexicon

By integrating 11 established audit tools and UGC, the study created a structured lexicon and demonstrates how traditional audit methodologies can be adapted to assess the presence and condition of park services for children through online reviews. The strong performance metrics of the lexicon validate its effectiveness in capturing these perceptions from UGC, the significant differences in perception found between most service pairs validate the classification of park features related to children. This provides a reliable framework for understanding the provision and performance of these services. The inclusion of three indicators directly derived from user reviews―amusement rides, book and reading facilities, and mosquitoes―highlights the potential of UGC to capture aspects that traditional assessment tools may overlook.

Although the lexicon was developed under Chinese contexts, its construction methodology and modular structure (9 dimensions, 44 indicators) have universal applicability and methodological reference in other contexts and languages.

5.2 Provision and performance of park services: insights from user perception

5.2.1 Provision differences of park services

Statistical analysis results showed distinct distribution patterns of perception among the 9 services, which uncovered differences in service provision across parks.

The normal distribution of Access, Nature, and Hygiene perceptions across parks indicates consistent awareness of these features, suggesting that they are fundamental aspects of park experiences and are widely provided across parks. Conversely, the non-normal distributions observed in other services imply inconsistent awareness among users, which may reflect uneven provision of these services. In particular, the distribution characteristics of Education service perceptions suggest substantial differences in the provision of educational services across parks. The contrast of coefficient of variation between Education and Access services suggests that educational features are highly park-specific and may depend on specialized programming or infrastructure, whereas considerations for Access are universally acknowledged regardless of park type or location. The highest satisfaction for Education services further indicates that despite limited provision, this resource-dependent service receives positive feedback from users who experience it, reflecting potential needs. Additionally, the largest standard deviations in perception frequency of Nature and Play services, along with their high mean perception, suggest that although these services are highly perceived by users, their quality, quantity, and design differ significantly between parks, creating diverse user experiences.

Notably, the differences in public perception between Education and Safety services revealed the distinction between low provision and low perception. While Education and Safety services exhibit the lowest mean perception frequency, their distribution characteristics differ significantly. Safety services show the smallest standard deviation, indicating consistently low perception across all parks. Conversely, Education services display high coefficients of variation and skewness, with discrete high perception values in certain parks, suggesting substantial variation in provision. This phenomenon suggests the presence of a “supply masking effect”: low mentions do not necessarily indicate low interest but may result from limited supply, which restricts users’ opportunities for experience and discussion.

5.2.2 Performance of services and priority improvements

The BERT sentiment analysis revealed the performance of services for children’s well-being. Education services received the highest satisfaction scores, which contrasts with their relatively low supply levels. The discrepancies in satisfaction for Hygiene and Safety services reflect significant quality variations across parks. Using the sentiment scores of indicators, this study further identified short-comings in Shanghai parks regarding support for children’s cycling activities. These detailed findings provide specific directions for service upgrades.

Moreover, the IPA pinpointed parks that exhibit relatively high perceptions but low satisfaction levels, clearly outlining the targets for priority improvements. This approach provides a more nuanced understanding of park service performance from users’ subjective perspectives, addressing limitations of traditional audit tools that may not accurately compare the performance of parks of different scales in supporting children’s well-being (Meng and Wang, 2022). Tracing negative feedback can further pinpoint specific optimization directions. For instance, the need for urgent improvements in Safety services at Park 7 arises from slippery metal surfaces, while the layout of Park 1’s play areas prevents effective adult supervision, increasing risks of conflict and safety issues for children.

This problem tracing analysis transcends simple satisfaction evaluations, offering concrete improvement pathways for park managers.

5.2.3 The dilemma of safety management: conflicting policies and user experience

Beyond performance assessment, analysis of negative comments reveals critical insights into management approaches, identifying a conflict between safety management practices and user expectations in Shanghai parks. Restrictive measures implemented by park managers for safety reasons―such as prohibiting strollers and children’s bicycles―have become significant sources of user dissatisfaction. Such simple prohibitions and restrictions to control risks undermine the core functions and appeal of parks as spaces for children’s activities (Brussoni et al., 2012).

Notably, the significance of safety services is highlighted through its synergistic effects with play services (as shown in the PDP analysis). This suggests that effective safety management should be supportive rather than restrictive, integrated into service design rather than imposed as external controls. This requires park managers to rethink the essence of safety. Truly high-quality park safety services should maximize support for diverse activities for children while ensuring reasonable risk control (Dodd and Lester, 2021; Jidovtseff et al., 2022), rather than simply evading responsibility through restrictions and prohibitions.

Furthermore, this restrictive management approach may reflect cultural differences in park governance, where Chinese parks’ restrictive policies may stem from considerations of order maintenance and potential conflict prevention, contrasting with Western approaches that generally emphasize facilitating children’s mobility and providing supportive infrastructure rather than imposing usage restrictions. This potentially reflects different understandings of public space use and risk management across cultural contexts.

5.3 The impact of services on park perception of children’s well-being

5.3.1 Beyond mention frequency: the significance of safety services

By employing relative residuals and three regression models, this study suggested that Safety, Comfort, Nature, and Play services will influence the perception of children’s well-being in parks. Safety service exerted the strongest influence despite its lowest mention frequency.

This finding highlights the importance of safety service, which aligns with prior research emphasizing that safety perception is a key factor encouraging children’s activities in parks (Bao et al., 2023). Notably, safety was also the lowest-scoring service in satisfaction, suggesting that users tend to comment primarily when they perceive problems. Therefore, even though safety features were mentioned less frequently than other services, these mentions often reflected notable concerns.

This phenomenon may explain why even infrequent mentions of safety can have a substantial impact. In contrast, Access services had the highest perception frequency, but they showed no significant impact. This finding challenges the common assumption that features frequently mentioned in reviews are necessarily the primary concerns of users, suggesting that the relationship between mention frequency and service importance is more complex than previously thought.

5.3.2 Interaction effects in park perception: play as a core service

While regression models identified Safety as the most influential service, the SVR model and partial dependence analysis revealed complex non-linear relationships and interaction effects between services, highlighting Play’s distinct role in influencing perception of children’s well-being.

Although Play showed the lowest coefficients among the four services in regression models, univariate PDPs revealed its distinctive nonlinear influence pattern: Play showed minimal impact until reaching a threshold, after which its influence increased substantially. This threshold effect suggests that play facilities must reach a certain quality or quantity before significantly influencing perception. The bivariate PDPs further reinforced this finding, revealing Play’s dominant role at higher levels across all service pairs.

The Play-Nature interaction showed that Nature dominated perception in parks with insufficient play facilities; however, when play features were adequate, users primarily perceived Play. Similarly, the Play-Comfort relationship showed that high Play values dominated regardless of Comfort levels, indicating that engaging play opportunities may compensate for comfort deficiencies. These findings highlight play’s core value from a user perspective, aligning with research suggesting that play facilities attract children to actively engage with urban parks (Bao et al., 2023). The emergence of Play as particularly influential at higher levels suggests that substantial investments in play environments may yield disproportionate benefits for children’s well-being perception, especially considering play’s fundamental role in childhood development (Freeman et al., 2021; Herrington and Brussoni, 2015).

Notably, the bivariate PDPs revealed important interaction effects that should inform integrated design approaches. The Play-Nature interaction demonstrated a compensatory relationship at lower values but synergistic effects at higher levels. This suggests that while either well-developed play facilities or natural elements can independently enhance park perception to some extent, their combination produces effects greater than the sum of their parts. This finding supports integrated design approaches that combine natural elements with play opportunities (Herrington and Brussoni, 2015). The Play-Safety interaction exhibited consistent synergistic effects, with each service enhancing the other’s impact. This relationship may suggest that users perceive safety features while commenting on play facilities. Given that safety services received the lowest satisfaction ratings while showing strong interaction with Play, improving safety measures in play areas appears particularly promising for enhancing park experiences. The depression observed in the Play-Comfort interaction (where moderate play facilities combined with low comfort features produce particularly poor outcomes) indicates that imbalanced service development may produce suboptimal effects, highlighting the importance of coordinated approaches to service provision.

These non-linear relationships and interaction effects provide potential insight into understanding why Play-Nature and Play-Comfort showed no significant differences in perception frequency, and suggest that park design should consider not only which services to include but also their relative development levels and combinations. The synergistic effects observed between certain service pairs suggest that strategic combinations of services may optimize experience outcomes with limited resources, offering a more efficient approach to park design than simply maximizing individual service quality.

6 Conclusions and recommendations

Addressing the insufficient attention to children in current UGC-based urban parks research, this study develops a reliable framework that integrates established audit tools and user reviews, and explores the potential of leveraging UGC to evaluate park services for children.

Analysis of variation in mention frequencies across parks revealed substantial differences in the provision of Education services, with their highest sentiment scores indicating potential needs for enhancing park experiences. Sentiment analysis and IPA further uncovered performance disparities, identified key deficiencies in Shanghai’s park services, and uncovered a fundamental conflict between safety management practices and children’s needs for outdoor activity.

Furthermore, the study revealed that Safety, Nature, Comfort, and Play services influence perceptions of children’s wellbeing in parks, capturing non-linear relationships and synergistic effects among these services using machine learning models.

These findings provide practical guidance for park optimization. Key recommendations include:

● Prioritizing the provision and improvement of play facilities, particularly in parks already rich in natural elements.

● Adopting integrated design approaches―such as embedding natural features into play areas and enhancing safety features without restricting activity.

● Developing the educational service potential in parks with suitable conditions.

● Transitioning from restrictive to supportive safety management, for example, by creating dedicated spaces for children’s cycling and skateboarding to control risks through spatial separation rather than prohibition.

Beyond the empirical findings, this study also contributes to UGC analysis methodology. The study identified the distinction between low perception and low supply, both of which can manifest as low mention frequencies. Additionally, the analysis revealed the latent significance of services that are not frequently mentioned, underscoring the limitations of UGC approaches relying solely on explicit feature extraction. These insights emphasize the complexity of user perception and the need for multi-dimensional approaches in UGC research.

The study demonstrates the feasibility of using UGC to investigate park services for children and highlights its potential in child-friendly environmental studies. While representational bias in UGC means it may not fully reflect the perspectives of children or caregivers, the abundant, long-term user experiences documented by park visitors can still offer a reliable complement compared to momentary expert audits. Importantly, this study does not interpret UGC sentiment as a direct indicator of absolute service quality. Instead, it uses inter-park sentiment variation to detect relatively underperforming parks. Nevertheless, inherent biases in UGC remain. Therefore, future assessments should interpret UGC-based findings cautiously and ideally supplement them with alternative data sources.

In addition to these concerns, the study acknowledges several other limitations. It does not explore temporal trends or shifts in sentiment, compare different park types, or analyze spatial patterns in service perception. Moreover, as the study is situated in Shanghai, its findings may be shaped by local cultural and societal factors, and therefore caution is needed when generalizing to other contexts.

Future research could integrate multiple data sources, including direct input from children, to provide more comprehensive insights; examine service disparities across various park typologies; and conduct longitudinal and spatial analyses to uncover temporal changes and potential equity concerns in service distribution; and expand the geographical scope to include diverse urban contexts, allowing for cross-cultural comparisons and validation of the proposed framework.

References

[1] Aarts, M.-J., de Vries, S.I., van Oers, H.A.M., Schuit, A.J., 2012. Outdoor play among children in relation to neighborhood characteristics: a cross-sectional neighborhood observation study. Int. J. Behav. Nutr. Phys. Activ. 9, 98.

[2] Amoly, E., Dadvand, P., Forns, J., López-Vicente, M., Basagaña, X., Julvez, J., Alvarez-Pedrerol, M., Nieuwenhuijsen, M.J., Sunyer, J., 2014. Green and blue spaces and behavioral development in Barcelona schoolchildren: the BREATHE project. Environ. Health Perspect. 122, 1351–1358.

[3] Andal, A.G., 2023. Children and urban vitalism amidst transitions: perspectives from slum-dwelling children in the Philippines. Cities 135, 104221.

[4] Bao, Y., Gao, M., Luo, D., Zhou, X., 2023. Urban Parks-A catalyst for activities! the effect of the perceived characteristics of the urban park environment on children’s physical activity levels. Forests 14, 423.

[5] Bi, J., Liu, Y., Fan, Z., Zhang, J., 2019. Wisdom of crowds: conducting importance-performance analysis (IPA) through online reviews. Tour. Manag. 70, 460–478.

[6] Bird, M.E., Datta, G.D., van Hulst, A., Kestens, Y., Barnett, T.A., 2015. A reliability assessment of a direct-observation park evaluation tool: the parks, activity and recreation among kids (PARK) tool. BMC Public Health 15, 906.

[7] Brussoni, M., Olsen, L.L., Pike, I., Sleet, D.A., 2012. Risky play and children’s safety: balancing priorities for optimal child development. Int. J. Environ. Res. Publ. Health 9, 3134–3148.

[8] Chen, S., Christensen, K.M., Li, S., 2019. A comparison of park access with park need for children: a case study in cache county, Utah. Landsc. Urban Plann. 187, 119–128.

[9] Chen, Y., Hong, C., Yang, Y., Li, J., Wang, Y., Zheng, T., Zhang, Y., Shao, F., De Meo, I., 2024. Mining social media data to capture urban park visitors’ perception of cultural ecosystem services and landscape factors. Forests 15, 213.

[10] Cheng, Y., Zhang, J., Wei, W., Zhao, B., 2021. Effects of urban parks on residents’ expressed happiness before and during the COVID-19 pandemic. Landsc. Urban Plann. 212, 104118.

[11] Christian, H., Zubrick, S.R., Foster, S., Giles-Corti, B., Bull, F., Wood, L., Knuiman, M., Brinkman, S., Houghton, S., Boruff, B., 2015. The influence of the neighborhood physical environment on early child health and development: a review and call for research. Health Place 33, 25–36.

[12] Cohen, D.A., Han, B., Williamson, S., Nagel, C., McKenzie, T.L., Evenson, K.R., Harnik, P., 2020. Playground features and physical activity in U.S. neighborhood parks. Prev. Med. 131, 105945.

[13] Crawford, D., Timperio, A., Giles-Corti, B., Ball, K., Hume, C., Roberts, R., Andrianopoulos, N., Salmon, J., 2008. Do features of public open spaces vary according to neighbourhood socioeconomic status? Health Place 14, 889–893.

[14] Curtis, D.S., Rigolon, A., Schmalz, D.L., Brown, B.B., 2022. Policy and environmental predictors of park visits during the first months of the COVID-19 pandemic: getting out while staying in. Environ. Behav. 54, 487–515.

[15] DeBate, R.D., Koby, E.J., Looney, T.E., Trainor, J.K., Zwald, M.L., Bryant, C.A., McDermott, R.J., 2011. Utility of the physical activity resource assessment for child-centric physical activity intervention planning in two urban neighborhoods. J. Community Health 36, 132–140.

[16] Devlin, J., Chang, M.-W., Lee, K., Toutanova, K., 2019. BERT: pretraining of deep bidirectional transformers for language understanding. In: 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Naacl Hlt 2019), Vol. 1. Presented at the Conference of the north-american-chapter of the association-for-computational-linguistics - Human Language Technologies (NAACL-HLT). Assoc Computational Linguistics-Acl, Stroudsburg, pp. 4171–4186.

[17] Dodd, H.F., Lester, K.J., 2021. Adventurous play as a mechanism for reducing risk for childhood anxiety: a conceptual model. Clin. Child Fam. Psychol. Rev. 24, 164–181.

[18] Donahue, M.L., Keeler, B.L., Wood, S.A., Fisher, D.M., Hamstead, Z.A., McPhearson, T., 2018. Using social media to understand drivers of urban park visitation in the twin cities, MN. Landsc. Urban Plann. 175, 1–10.

[19] Dong, X., Geng, L., 2023. Nature deficit and mental health among adolescents: a perspectives of conservation of resources theory. J. Environ. Psychol. 87, 101995.

[20] Dunkel, A., 2015. Visualizing the perceived environment using crowdsourced photo geodata. Landsc. Urban Plann. 142, 173–186.

[21] Freeman, N., Gage, R., Chambers, T., Blaschke, P., Cook, H., Stanley, J., Pearson, A., Smith, M., Barr, M., Signal, L., 2021. Where do the children play? An objective analysis of children’s use of green space. Health Promot. Int. 36, 846–853.

[22] Gao, M., Cheng, X., Bao, Y., Zhou, X., 2024. Beyond green: unveiling the impact of urban park quality and greenery on children’s physical activity. Environ. Plan. B Urban Anal. City Sci. 52 (7), 1717–1734.

[23] Garau, C., Annunziata, A., 2019. Smart city governance and children’s agency: an assessment of the green infrastructure impact on children’s activities in cagliari (Italy) with the tool “Opportunities for Children in Urban Spaces (OCUS).”. Sustainability 11, 4848.

[24] General Office of the Shanghai Municipal People’s Government (GOSMPG), 2022. Implementation plan for the construction of child-friendly cities in Shanghai (IPCCFCS).

[25] Guerrero, P., Møller, M.S., Olafsson, A.S., Snizek, B., 2016. Revealing cultural ecosystem services through Instagram images: the potential of social media volunteered geographic information for urban green infrastructure planning and governance. Urban Planning 1, 1–17.

[26] Gustat, J., Anderson, C., Slater, S., 2019. Development and testing of a brief play space audit tool. J. Phys. Activ. Health 17, 1–8.

[27] Hausmann, A., Toivonen, T., Fink, C., Heikinheimo, V., Kulkarni, R., Tenkanen, H., Di Minin, E., 2020. Understanding sentiment of national park visitors from social media data. People and Nature 2, 750–760.

[28] He, H., Sun, R., Li, J., Li, W., 2023. Urban landscape and climate affect residents’ sentiments based on big data. Appl. Geogr. 152, 102902.

[29] Heikinheimo, V., Tenkanen, H., Bergroth, C., Järv, O., Hiippala, T., Toivonen, T., 2020. Understanding the use of urban green spaces from user-generated geographic information. Landsc. Urban Plann. 201, 103845.

[30] Herrington, S., Brussoni, M., 2015. Beyond physical activity: the importance of play and nature-based play spaces for children’s health and development. Current Obesity Reports 4, 477–483.

[31] Huai, S., Liu, S., Zheng, T., Van de Voorde, T., 2023. Are social media data and survey data consistent in measuring park visitation, park satisfaction, and their influencing factors? A case study in Shanghai. Urban For. Urban Green. 81, 127869.

[32] Huai, S., Van de Voorde, T., 2022. Which environmental features contribute to positive and negative perceptions of urban parks? A cross-cultural comparison using online reviews and natural language processing methods. Landsc. Urban Plann. 218, 104307.

[33] Istiani, N.F.F., Alkadri, M.F., van Nes, A., Susanto, D., 2023. Investigating the spatial network of playgrounds during covid-19 based on a space syntax analysis case study: 10 playgrounds in delft, the Netherlands. Cogent Social Sciences 9, 2163754.

[34] Jenkins, G.R., Yuen, H.K., Rose, E.J., Maher, A.I., Gregory, K.C., Cotton, M.E., 2015. Disparities in quality of park play spaces between two cities with diverse income and race/ethnicity composition: a pilot study. Int. J. Environ. Res. Publ. Health 12, 8009–8022.

[35] Jidovtseff, B., Pirard, F., Martin, A., McCrorie, P., Vidal, A., Pools, E., 2022. Parental assessment of benefits and of dangers determines children’s permission to play outdoors. Int. J. Environ. Res. Publ. Health 19, 11467.

[36] Kaczynski, A.T., Stanis, S.A.W., Besenyi, G.M., 2014. Community park audit tool.

[37] Kaczynski, A.T., Wilhelm Stanis, S.A., Besenyi, G.M., 2012. Development and testing of a community stakeholder park audit tool. Am. J. Prev. Med. 42, 242–249.

[38] Koblet, O., Purves, R.S., 2020. From online texts to landscape character assessment: collecting and analysing first-person landscape perception computationally. Landsc. Urban Plann. 197, 103757.

[39] Komossa, F., Wartmann, F.M., Kienast, F., Verburg, P.H., 2020. Comparing outdoor recreation preferences in peri-urban landscapes using different data gathering methods. Landsc. Urban Plann. 199, 103796.

[40] Li, J., Fu, J., Gao, J., Zhou, R., Wang, K., Zhou, K., 2023. Effects of the spatial patterns of urban parks on public satisfaction: evidence from Shanghai, China. Landsc. Ecol. 38, 1265–1277.

[41] Liu, W., Hu, X., Song, Z., Yuan, X., 2023. Identifying the integrated visual characteristics of greenway landscape: a focus on human perception. Sustain. Cities Soc. 99, 104937.

[42] Loukaitou-Sideris, A., Sideris, A., 2009. What brings children to the park? Analysis and measurement of the variables affecting children’s use of parks. J. Am. Plann. Assoc. 76, 89–107.

[43] Louv, R., 2005. Last child in the woods: saving our children from. Nature-Deficit Disorder. Algonquin Books, Chapel Hill, NC.

[44] Marti, P., Serrano-Estrada, L., Nolasco-Cirugeda, A., 2019. Social media data: challenges, opportunities and limitations in urban studies. Comput. Environ. Urban Syst. 74, 161–174.

[45] Martilla, J.A., James, J.C., 1977. Importance-performance analysis. J. Market. 41, 77–79.

[46] Matzler, K., Bailom, F., Hinterhuber, H.H., Renzl, B., Pichler, J., 2004. The asymmetric relationship between attribute-level performance and overall customer satisfaction: a reconsideration of the importance–performance analysis. Ind. Mark. Manag. 33, 271–277.

[47] Meng, X., Wang, M., 2022. Comparative review of environmental audit tools for public open spaces from the perspective of children’s activity. Int. J. Environ. Res. Publ. Health 19, 13514.

[48] Ministry of Housing and Urban-Rural Development (MHURD), National Development and Reform Commission (NDRC), National Working Committee on Children and Women, 2023. Guidelines for the Construction of Child-Friendly Urban Spaces (GCCFUS).

[49] National Bureau of Statistics of China (NBSC), UNICEF China, UNFPA China, 2023. What the 2020 census can tell us about children in China: facts and figures.

[50] National Development and Reform Commission (NDRC), National Working Committee on Children and Women, Ministry of Housing and Urban-Rural Development (MHURD), Publicity Department of China, Cyberspace Administration of China, Ministry of Education, Ministry of Public Security, Ministry of Civil Affairs, 2021. Guidance on Promoting the Construction of Child-Friendly Cities (GPCCFC).

[51] Park, K., 2017. Psychological park accessibility: a systematic literature review of perceptual components affecting park use. Landsc. Res. 42, 508–520.

[52] Pham, T.-T.-H., Labbé, D., Lachapelle, U., Pelletier, É., 2019. Perception of park access and park use amongst youth in hanoi: how cultural and local context matters. Landsc. Urban Plann. 189, 156–165.

[53] Pokhriyal, N., Valentino, B.A., Vosoughi, S., 2023. Quantifying participation biases on social media. EPJ Data Science 12, 1–20.

[54] Rigolon, A., 2017. Parks and young people: an environmental justice study of park proximity, acreage, and quality in Denver, Colorado. Landsc. Urban Plann. 165, 73–83.

[55] Rigolon, A., Németh, J., 2018. A QUality INdex of parks for youth (QUINPY): evaluating urban parks through geographic information systems. Environ. Plan. B Urban Anal. City Sci. 45, 275–294.

[56] Rui, J., 2023. Exploring the association between the settlement environment and residents’ positive sentiments in urban villages and formal settlements in Shenzhen. Sustain. Cities Soc. 98, 104851.

[57] Saelens, B.E., Frank, L.D., Auffrey, C., Whitaker, R.C., Burdette, H.L., Colabianchi, N., 2006. Measuring physical environments of parks and playgrounds: EAPRS instrument development and inter-rater reliability. J. Phys. Activ. Health 3, S190–S207.

[58] Sessions, C., Wood, S.A., Rabotyagov, S., Fisher, D.M., 2016. Measuring recreational visitation at US national parks with crowd sourced photographs. J. Environ. Manag. 183, 703–711.

[59] Shanghai Landscaping & City Appearance Administrative Bureau (SLCAAB), 2024. Directory of parks (Green Spaces).

[60] Shanghai Municipal Development and Reform Commission (SMDRC), 2023. Three-Year Action Plan for Child-Friendly City Construction (TYAPCFCC) (2023-2025).

[61] Shelton, T., Poorthuis, A., Zook, M., 2015. Social media and the city: rethinking urban socio-spatial inequality using user-generated geographic information. Landsc. Urban Plann. 142, 198–211.

[62] Shi, D., Zhang, B., Yi, B., 2022. How is tourist destination perception formed? Exploratory research based on text mining. Tour. Trib. 37, 68–82.

[63] Sikorska, D., Łaszkiewicz, E., Krauze, K., Sikorski, P., 2020. The role of informal green spaces in reducing inequalities in urban green space availability to children and seniors. Environ. Sci. Pol. 108, 144–154.

[64] Song, Y., Wang, R., Fernandez, J., Li, D., 2021. Investigating sense of place of the Las Vegas strip using online reviews and machine learning approaches. Landsc. Urban Plann. 205, 103956.

[65] Standardization Administration of the People’s Republic of China (SAC), 2011. Safety for outdoor body-building equipment―General requirements (GB 19272-2011).

[66] Standardization Administration of the People’s Republic of China (SAC), 2012. No power type of amusement devices - terminology (GB/T 28622-2012).

[67] Standardization Administration of the People’s Republic of China (SAC), 2017. Amusement devices terminology (GB/T 20306-2017).

[68] Standardization Administration of the People’s Republic of China (SAC), 2024. Sport Information Classification and Code - Part 5: Sport Venue Code (GB/T 34308.5-2024).

[69] Sun, J., 2020. Jieba: chinese words segmentation utilities. Version 0.42.1.

[70] Tang, Z., Zhao, Y., Fu, M., Wang, Y., Xue, J., 2023. Which factors influence public perceptions of urban attractions? ― a comparative study. Ecol. Indic. 154, 110541.

[71] Tenkanen, H., Di Minin, E., Heikinheimo, V., Hausmann, A., Herbst, M., Kajala, L., Toivonen, T., 2017. Instagram, flickr, or Twitter: assessing the usability of social media data for visitor monitoring in protected areas. Sci. Rep. 7, 17615.

[72] UNICEF, 2012. The State of the World’s Children 2012: Children in an Urban World (United Nations).

[73] Veitch, J., Ball, K., Flowers, E., Deforche, B., Timperio, A., 2021. Children’s ratings of park features that encourage park visitation, physical activity and social interaction. Urban For. Urban Green. 58, 126963.

[74] Veitch, J., Salmon, J., Ball, K., Crawford, D., Timperio, A., 2013. Do features of public open spaces vary between urban and rural areas? Prev. Med. 56, 107–111.

[75] Wan, C., Shen, G.Q., Choi, S., 2021. Eliciting users’ preferences and values in urban parks: evidence from analyzing social media data from Hong Kong. Urban For. Urban Green. 62, 127172.

[76] Wang, Z., Miao, Y., Xu, M., Zhu, Z., Qureshi, S., Chang, Q., 2021a. Revealing the differences of urban parks’ services to human wellbeing based upon social media data. Urban For. Urban Green. 63, 127233.

[77] Wang, Z., Zhu, Z., Xu, M., Qureshi, S., 2021b. Fine-grained assessment of greenspace satisfaction at regional scale using content analysis of social media and machine learning. Sci. Total Environ. 776, 145908.

[78] Wankhade, M., Rao, A.C.S., Kulkarni, C., 2022. A survey on sentiment analysis methods, applications, and challenges. Artif. Intell. Rev. 55, 5731–5780.

[79] Wei, D., Liu, M., Grekousis, G., Wang, Y., Lu, Y., 2023. User-generated content affects urban park use: analysis of direct and moderating effects. Urban For. Urban Green. 90, 128158.

[80] Wilkins, E.J., Van Berkel, D., Zhang, H., Dorning, M.A., Beck, S.M., Smith, J.W., 2022. Promises and pitfalls of using computer vision to make inferences about landscape preferences: evidence from an urban-proximate park system. Landsc. Urban Plann. 219, 104315.

[81] Woolley, H., Lowe, A., 2013. Exploring the relationship between design approach and play value of outdoor play spaces. Landsc. Res. 38, 53–74.

[82] Yan, Z., Guo, X., Zhao, Z., Tang, L., 2024. Achieving fine-grained urban flood perception and spatio-temporal evolution analysis based on social media. Sustain. Cities Soc. 101, 105077.

[83] Yang, J., Mu, L., Grigsby-Toussaint, D.S., 2024. Public perceptions and expectations of urban park environments on children’s play: a GIS and text mining analysis on children’s activity in Atlanta’s parks using social media data. Environ. Plan. B Urban Anal. City Sci.

[84] Zhang, S., Zhou, W., 2018. Recreational visits to urban parks and factors affecting park visits: evidence from geotagged social media data. Landsc. Urban Plann. 180, 27–35.

[85] Zhao, X., Lu, Y., Huang, W., Lin, G., 2024. Assessing and interpreting perceived park accessibility, usability and attractiveness through texts and images from social media. Sustain. Cities Soc. 112, 105619.

[86] Zhao, Y., Ruan, P., Liu, X., Shan, X., 2020. Study on user satisfaction evaluation based on comment. Manag. Rev. 32, 179–189.

[87] Zhou, S., Wang, H., Li, D., Ng, S.T., Wei, R., Zhao, Y., Zhou, Y., 2024. Revealing public attitudes toward mobile cabin hospitals during Covid-19 pandemic: sentiment and topic analyses using social media data in China. Sustain. Cities Soc. 107, 105440.

RIGHTS & PERMISSIONS

2095-2635/2025 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd.