Introduction

Emotion is a subjective psychological state influenced by internal and external stimuli, characterized by changes in physiological, cognitive, and behavioral aspects1. It significantly impacts individual decision-making, social interactions, and well-being, as well as public health, community cohesion, and quality of life at the societal level. With the acceleration of urbanization, the urban environment, particularly green spaces, as a crucial component of the urban ecosystem, has been shown to play a significant role in emotion regulation and mental health2. According to the “Pleasure from Nature Theory"3 natural environments restore emotional balance, with various types of green spaces affecting emotions and health differently. In the context of globalization, the emotional impacts of urban green spaces may differ according to cultural variations. “Environmental psychology studies"4 have identified cultural and ethnic differences in landscape preferences. “Affective geography"5 further reveals how spaces shape emotional experiences, emphasizing that the design and use of parks as public spaces may carry different emotional meanings across cultures. Therefore, this study not only contributes to optimizing urban planning and green space design while providing a scientific foundation for promoting social inclusiveness, enhancing public health, and improving the well-being of urban residents.

As integral parts of urban environments, urban parks are closely connected to emotions and play a series of important roles: Firstly, as urban public green spaces, they effectively promote physical and psychological health, alleviate tension and negative emotions, and improve social environmental quality6. Secondly, the recreational and sports facilities provided by urban parks meet people’s needs for leisure and fitness, increasing the time and frequency of outdoor activities such as biking and walking7. Thirdly, urban parks also serve as spaces for cultural dissemination, offering platforms for spreading civilization, conducting scientific education, displaying cultural heritage, and providing various environmental, aesthetic, and recreational benefits8. As a whole, urban parks exert a profound and positive impact on the emotional well-being of city residents by promoting physical and mental health, providing opportunities for recreation and leisure, and disseminating cultural values.

London’s parks and gardens are hailed as the green lungs of the Greater London area, effectively improving the urban environment and increasing greenery and public spaces, providing citizens with more outdoor and recreational venues9. In contrast, Shanghai, as one of China’s economic centers, has seen a reduction in urban green spaces due to rapid urbanization10, with original green spaces reduced or lost due to land development and construction, triggering a series of urban problems. To alleviate environmental pressure and enhance visitors’ positive emotions, Shanghai has introduced several policies and measures to increase green spaces and improve quality of life.

This study aims to explore the relationship between emotions of urban visitors from different cultural backgrounds and urban park environments, identifying key factors in urban park environments that affect visitors’ emotions. This study uses online review data from social media to analyze the positive emotions of urban visitors. Through cross-cultural comparisons, this research can delve into the similarities and differences in positive emotions and preferences of urban visitors from different cultural backgrounds. We selected Shanghai and London as regions for cross-cultural comparison. Firstly, online review data were processed using Natural Language Processing (NLP) techniques to quantify the positive emotional well-being values of urban park visitors. Secondly, we used the LightGBM regression model to explore the nonlinear relationship between urban park environments and visitors’ positive emotions and identified the main factors affecting positive emotions within urban parks. Finally, spatial autocorrelation analysis was conducted, and an MGWR model was constructed to identify the scale effects of different variables and reveal the causes of spatial heterogeneity.

Literature review

Research on emotions and cross-cultural differences in urban spaces

In recent years, the emotional research within urban spaces has garnered widespread attention. Scholars have explored how different types of urban spaces, such as urban forests2, green areas9, parks10, public spaces11, waterfront areas12, and communities, affect the emotions of urban visitors differently. Studies have noted significant differences in emotional scores inside and outside parks13. Urban spaces influence human emotional responses through the arrangement of spatial sequences and the transition of scenographic sequences14. “Situational emotion theory"5 posits that emotional responses are shaped not only by the physical environment but also by the cultural context. For instance, in Western cultures, visitors often focus on the relaxation benefits of parks, whereas in Eastern cultures, parks are seen as spaces for socializing and family gatherings.

Research on urban park environments and visitor emotions

Research has long explored the impact of physical and other attributes of urban park environments on visitor emotions. Certain environmental features2,8,15, such as park area, water bodies, greenery, and facilities, have been identified as closely related to visitor emotions. Scholars have noted that the frequency of perception of water bodies and colors, as well as crowd frequency, are positively correlated with users’ perceived emotions8. Another aspect discussed is how other attributes of urban parks16,17, such as distance, type, and safety, affect emotions. For more comprehensive studies, researchers have employed various methods to collect well-being data. Self-report measures18, questionnaires, and demographic surveys19 are the most commonly used approaches. Additionally, smartphone apps and wearable sensors, combined with GPS ___location data12, are increasingly being utilized.

Despite ongoing research accumulation, related studies remain limited. Firstly, existing research on urban parks mainly focuses on how the internal environment, such as vegetation, roads, and related facilities, affects visitor emotions. while studies on how both external and internal environments of parks collectively influence visitor emotions are relatively scarce. Secondly, although previous research has explored the use and preferences of different demographic groups and different cultures for urban park environments, there is still a lack of research on different groups in diverse socio-cultural backgrounds, especially given the increasingly common context of globalized cities. Thirdly, current research primarily uses residents’ self-reported psychological stress or survey questionnaires to explore the relationship between urban park environments and visitor emotions. The challenge lies in the potential interference from excessive subjective agency.

Data and methodology

Study area

This study selected Shanghai, China, and London, UK, as empirical case studies with different cultural backgrounds (Fig. 1). Shanghai (121°50’E, 31°40’N), one of the top ten cities in the world, had a permanent population of about 24.76 million and an urban green space area of 1646 square kilometers, with an average green space area per capita of 8.8 square meters20. London (0°7’39"W, 51°30’26"N), the capital and largest city of the UK, covered an area of 1,577 square kilometers with a population exceeding 9 million, and had 675 square kilometers of open space, with an average green space per capita of 18.98 square meters21. The study areas selected were the central urban area of Shanghai and Inner London, representing different socio-cultural backgrounds from Asia and Europe. Both cities were important global economic centers and multicultural hubs, with population densities of 23,092 people/km² in central Shanghai and 5,701 people/km² in London20,22. The urban planning departments of both cities placed a high priority on green space development; Shanghai aimed to reach a per capita park area of 13.1 square meters by 2035, while London had entered a mature phase of refined management of urban green spaces. By studying parks of different scales in these cities(with large parks defined as those larger than 0.2 km², medium parks as 0.05–0.2 km², and small parks as less than 0.05 km²), this research aimed to explore the factors influencing positive emotions and usage preferences of park visitors across different cultural contexts, to reveal potential geographic disparities, and to provide recommendations for urban planning and park management.

Fig. 1
figure 1

Study areas: (a) Shanghai (b) London (It was generated by ArcMap 10.6 software (https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview)).

Full size image

Research framework

The experimental design and process, as shown in Fig. 2, involved selecting Shanghai and London as two areas for cross-cultural comparison. Visitor positive emotion values were derived using NLP analysis of online text data collected from Sina Weibo and Twitter. Concurrently, environmental data related to parks were collected as independent variables, with visitor emotion values serving as dependent variables to construct regression models. The study employed correlation analysis, machine learning, Moran’s I index, and Multiscale Geographically Weighted Regression (MGWR) to quantify visitor emotion values and identify urban park environmental features that significantly influence visitor positive emotions. This research framework integrated urban visitors and park environments from different socio-cultural backgrounds, overcoming the limitations of traditional studies that focused on a single background for studying urban visitor emotions. The study identified the importance of various factors influencing positive emotions in parks through machine learning regression models. Furthermore, global Moran’s I was used to determine the spatial autocorrelation of all variables, and local Moran’s scatter plots visualized the geographical spatial autocorrelation of the dependent variables. Finally, the study used the MGWR model to geographically explore the promoting or inhibiting effects of park environmental elements on visitor positive emotions towards parks. By quantifying the emotions of visitors in Shanghai and London, representative cities from Asia and Europe, the study highlighted cultural differences in park use and evaluations of park environments. This comparative research advances park construction from cultural and perceptual dimensions to enhance the quality of urban green spaces and resident well-being.

Text data processing using NLP

Before conducting sentiment analysis, it is necessary to preprocess the texts to reduce biases caused by extraneous information in each tweet. Initially, we employed Jieba for data preprocessing of Chinese comments, which included data cleaning, tokenization, and stopword removal. The Natural Language Toolkit (NLTK) performed similar data preprocessing for English comments, including data cleaning, lowercasing, stemming (to identify the root of words), and stopword removal23. Non-text characters were also removed, including punctuation, tags, URLs, and numbers. Additionally, data such as official announcements, advertisements, and news were excluded. Subsequently, we filtered both Chinese and English textual content using ten keywords—park, green space, garden, tree, lake, plant, flower, architecture, landscape, and nature—to ensure relevance to urban parks. Further, tweets shorter than 5 or longer than 150 words were removed24, as too few or too many words are detrimental to emotional well-being analysis. Only English tweets were selected for analysis.

Natural Language Processing (NLP)25 methods can overcome limitations of data sources and transform vast amounts of unstructured and open-ended data into structured and systematized information, providing efficient solutions. Specifically, sentiment analysis can translate intangible emotions into concrete numerical scores26, allowing for a clear assessment of urban parks’ impact on visitors’ emotions. We conducted sentiment analysis on both Chinese and English texts using the Baidu Natural Language Processing (NLP) platform (https://ai.baidu.com/tech/NLP), quantifying visitor emotions. The Baidu NLP platform is a powerful text processing library that is based on pre-trained machine learning models and can achieve high-precision calculations in sentiment analysis tasks27. The sentiment scores range from low to high (between 0 and 1). Using the quantile method, we set thresholds to classify online reviews into three sentiment polarities: positive (0.67 < sentiment score ≤ 1), neutral (0.33 < sentiment score ≤ 0.67), and negative (0 < sentiment score ≤ 0.33). We obtained the sentiment scores for each online comment via their application programming interface (API). The average sentiment score of all comments for each park was used as a proxy for visitor emotions.

Fig. 2
figure 2

Research framework.

Full size image

Selection of explanatory variables

Emotions can be expressed through social media, and research has confirmed that social media data can be used to analyze the association between visitor emotions and factors such as nature and socio-economic conditions28,29. Online reviews of urban parks in Shanghai and London are sourced from Weibo and Twitter, respectively. Weibo2, with 599 million monthly active users, is the largest open social media platform in China, while Twitter16 is a significant global social media platform that facilitates cross-cultural communication and information connectivity across different cultures and languages worldwide. We utilized these two social media data platforms due to their broad user base and influence, and their application across various research fields. Social media data was obtained using the Python programming language (version 3.7) from the Weibo and Twitter APIs (https://m.weibo.cn, https://twitter.com/). Metadata includes geographic coordinates, user and post IDs, timestamps, and annotations such as text, image links, etc. This data covers the period from February 1, 2023, to August 1, 2023. Weibo data contains geo-tags, while Twitter data is collected through text searches using park names as keywords. We collected a total of 343,575 Weibo posts and 231,161 Twitter texts. After filtering out irrelevant texts, we obtained 6,932 Weibo texts and 4,592 Twitter texts.

To identify factors influencing positive emotions among urban park visitors, we collected data related to the urban park environment (Table 1), including administrative boundaries, Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Points of Interest (POI), Areas of Interest (AOI), urban park and city road data, and economic vitality data. This comprehensive data collection allows us to analyze and correlate various environmental and social factors with the emotional well-being of park visitors, thereby providing a more nuanced understanding of how urban green spaces impact well-being.

Table 1 The data sources and processing methods of all variables.
Full size table

Model architecture

LightGBM

LightGBM (Light Gradient Boosting Machine), developed by Microsoft on the basis of GBDT, has become one of the most effective machine learning algorithms. It introduces a histogram-based algorithm34 and utilizes the gradient boosting framework along with two new techniques: Gradient-based One-Side Sampling (GOSS) for handling large data instances and Exclusive Feature Bundling (EFB) for managing a large number of features or variables. Its efficacy lies in the innovative approach to tree-based learning. Unlike traditional gradient boosting algorithms, LightGBM adopts a leaf-wise (best-first) strategy that results in a more efficient and faster training process35, effectively reducing the number of nodes in trees, thus enhancing memory utilization and computational speed. The main features of LightGBM are GOSS and EFB, along with histogram and leaf-wise growth strategies. Given training data \(\:{\left\{\left({x}_{i},{y}_{i}\right)\right\}}_{i=1}^{N}\), where \(\:{x}_{i}\) are explanatory factors (such as NDVI, NDWI, accessibility, etc.), and \(\:{y}_{i}\) are the target values (i.e.,the emotional well-being values of visitors in parks). The objective function of LightGBM is defined as follows in Eqs. (1) and (2):

$$Obj = \mathop \sum \limits_{{i = 1}}^{n} l\left( {y_{i} ,\hat{y}_{i} } \right) + \mathop \sum \limits_{{k = 1}}^{K} {{\Omega }}\left( {f_{k} } \right)$$
(1)
$${{{\Omega}}}\left( {{f_k}} \right)=\gamma T+\frac{1}{2}\lambda \parallel \omega {\parallel ^2}$$
(2)

Another significant improvement of LightGBM is the simultaneous use of first-order and second-order derivative loss functions to prevent overfitting. Hence, the LightGBM algorithm can better handle the constraints of data scarcity.

Moran’s I test

Spatial autocorrelation describes the non-random spatial distribution of data across a geographical area. It encompasses global Moran’s I36, which assesses overall spatial correlation trends across a study area, and local Moran’s I, or Local Indicators of Spatial Association (LISA)37, which detects local clustering of similar values. In this study, both indices are used to analyze the spatial patterns and clustering of visitor emotions, helping to understand how these emotions are distributed and vary geographically. Meanwhile, we analysed the data using the Kriging interpolation method in ArcGIS 10.6. We selected the Gaussian model (Eq. (3)) as the variogram model due to its ability to effectively capture the spatial variability characteristics of the sentiment data. Additionally, we set the search radius to 300 meters28 to ensure that the spatial correlation of neighbouring areas was adequately considered during the interpolation process.

$$\gamma (h)={c_0}+c\left( {1 - \exp \left( { - \frac{{{h^2}}}{{{a^2}}}} \right)} \right)$$
(3)

Where: \(\:\gamma\:\left(h\right)\) is the semivariance at distance \(\:h\), representing the degree of spatial variability at that distance. \(\:{c}_{0}\) is the nugget effect, which represents the micro-scale variability or measurement error at zero distance. \(\:c\) is the sill, which represents the maximum semivariance when the variogram reaches its plateau, indicating the total variability in the data. \(\:a\) is the range, which represents the maximum distance over which spatial autocorrelation exists. Beyond this distance, the semivariance stabilizes. \(\:\:h\) is the spatial distance between two points.

Multiscale geographically weighted regression

MGWR (Multiscale Geographically Weighted Regression) is an innovative approach that relaxes the assumption that all modeled processes operate at the same spatial scale, allowing for the relationships between dependent and independent variables to vary across different spatial scales38. The model addresses this limitation by considering multiple bandwidths, thereby identifying scale effects of different variables’ impacts39. This study employs this model to investigate the spatial non-stationarity and scale effects between the emotional well-being values of urban park visitors and other factors, with the formula presented in Eq. (4):

$$y_{i} = \mathop \sum \limits_{{j = 1}}^{k} \beta _{{{\text{bw}}j}} \left( {u_{i} ,v_{i} } \right)x_{{ij}} + \varepsilon _{i}$$
(4)

In the equation, \(\:{y}_{i},{x}_{ij},{\epsilon\:}_{i}\) represent the dependent variable, independent variables, and random errors at spatial ___location\(\:\:i\), respectively;\(\:\:\left({u}_{i},{v}_{i}\right)\) denotes the spatial coordinates of point \(\:i\); \(\:j\) represents the number of independent variables;\(\:\:\beta\:\) is the regression coefficient at point \(\:i\). \(bwj\)represents the bandwidth used for the regression coefficient of the \(\:j\)th variable. Each regression coefficient \(\:{\beta\:}_{\text{b}\text{w}j}\) in MGWR is derived from local regressions and is specific to its bandwidth. MGWR continues to utilize kernel functions and bandwidth selection criteria. This paper employs the commonly used bisquare kernel function and the\(AICc\)criterion.

Fig. 3
figure 3

Distribution of Visitor Emotional well-being Values in Urban Parks: (a) Shanghai (b) London (It was generated by ArcMap 10.6 software (https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview)).

Full size image

Results

Emotional well-being values of visitors in urban parks in Shanghai and London

Figure 3 presents the distribution maps of visitor emotional well-being values in urban parks in Shanghai and London. Through these maps, we can clearly observe the spatial patterns of park emotional well-being values in both cities. By calculating the number of parks and the average emotional well-being values within each administrative district (Fig. 4 (a)(b)), as well as the average emotional well-being values of visitors in urban parks of different scales, it was found that the overall emotional well-being values of visitors in Shanghai’s parks were generally higher, and there was significant variability in emotional well-being values across different districts. Figure 4(a) shows that the highest emotional well-being values in Shanghai were found in the northern districts of Baoshan, Pudong, Hongkou, and Putuo, while the lowest emotional well-being values were in the southern districts of Minhang, Huangpu, and Jing’an. Figure 4(c) reveals that visitors in Shanghai preferred to express positive emotions in medium-sized parks, followed by large parks, whereas small parks were less likely to inspire positive emotions among visitors. In contrast, the overall emotional well-being values of visitors in London were slightly lower than those in Shanghai, and the emotional well-being values were more evenly distributed across different districts. Notably, unlike in Shanghai, where visitors more readily expressed positive emotions in medium-sized parks, in London, the emotional well-being values in large parks were lower than in medium and small parks, with small parks being the most likely to elicit positive emotions (Fig. 4(d)). This indicates differing preferences for park sizes between visitors in Shanghai and London. Additionally, Fig. 4(b) shows that the lowest emotional well-being values in London parks were more likely to occur in the northern districts of Hackney, Newham, and Islington, which typically host large urban parks. Conversely, the highest emotional well-being values were found in the southern districts of Lambeth, Greenwich, and Southwark, and parks south of the Thames River exhibited higher emotional well-being values.

Results of machine learning model

Before conducting the machine learning regression analysis, a correlation analysis was first carried out on the data to filter variables (see Appendix A1). The results indicated no significant correlation between road density and the emotional well-being values of visitors, leading to the exclusion of this variable. Numerous studies have demonstrated that nonlinear models possess robust feature processing capabilities and can deliver high predictive accuracy, particularly in scenarios involving complex data distributions or large sample sizes40. Two popular machine learning algorithms were selected,

Fig. 4
figure 4

Emotional well-being Values of Visitors in Different Administrative Districts and Parks of Various Scales.

Full size image

Random Forest Regression, and LightGBM Regression, to train and predict datasets from two cities. The datasets were divided into training sets (60–70%), validation sets (10–20%), and test sets (10–20%).In this study, hyperparameters included max depth, learning rate, gamma, minimum child weight, and estimators41. Table A1 displays the range of values for these parameters. For each combination of values, GridSearchCV42 was applied—a function that automatically iterates over multiple parameter combinations and returns the most suitable one—and performed five-fold cross-validation to select the optimal values to minimize error rates. The iteration process was automatically terminated when there was no further improvement in the model evaluation score, and the optimal combination of hyperparameters was obtained. The two machine learning regression models were then validated to assess which model yielded the best results, using R2, Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) to evaluate model performance. Models with lower RMSE and MAPE values indicated higher accuracy. The results showed that the LightGBM regression model performed exceptionally well (Table 2), and thus, it was selected for further analysis.

Table 2 Results of the LightGBM regression model.
Full size table

According to the results of the LightGBM regression model (Fig. 5), we first observed that in Shanghai, the importance of independent variables is densely clustered, with the lowest and highest values ranging from 12 to 21.6%. In contrast, in London, there is a larger spread in the importance values, ranging from 10 to 21.6%. In Shanghai, the top four independent variables affecting visitor emotions in urban parks are NDWI (21.6%), NDVI (18.2%), economic vitality (18%), and Accessibility (17.2%). These results indicate that water bodies play a crucial role in influencing the activities and emotions of visitors in Shanghai’s urban parks. However, in London, water bodies play only a moderate role. Additionally, NDVI values are typically associated with green landscapes, indicating that natural environments can enhance visitors’ positive emotions. However, NDVI does not significantly influence the positive emotions of visitors in London, which may be related to the city’s inherently high green space ratio. Furthermore, nighttime light data, which represents economic vitality, demonstrates that stronger nighttime activity levels and urban vibrancy also significantly impact the emotions of urban visitors within parks. Finally, the results suggest that accessibility plays a role in boosting the emotions of visitors to urban parks in Shanghai.

Fig. 5
figure 5

Ranking of the importance of independent variables affecting visitor emotions in urban parks in Shanghai and London (a. Shanghai b. London).

Full size image

In contrast, in London, the most important variable is accessbility (21.6%), indicating its significant influence on the emotions of urban park visitors. Similar to Shanghai, economic vitality ranks second (19.6%), though its impact on visitor emotions is relatively stronger in London. Parking lot density ranks third (18%), suggesting that the availability and infrastructure of parking facilities are strongly associated with visitors’ emotions in London’s parks. In comparison, parking lot density has a lower importance contribution in Shanghai. The fourth-ranked education density (17.3%) suggests that areas with higher concentrations of educational and cultural facilities have more public educational amenities, which also affect people’s emotions.

Spatial autocorrelation analysis

Global Moran’s\(\:\:I\) was conducted to verify the spatial autocorrelation of each variable. The results for the univariate global Moran’s I are presented in Table 3. All significance levels were at 0.001 (<0.05), passing the 5% significance test level. This indicates the presence of spatial autocorrelation within the distribution of variables. In Shanghai, the Moran’s\(\:\:I\) for variables such as Leisure density, park accessibility, parking lot density, and education density were 0.937, 0.935, 0.935, and 0.903, respectively, showing a strong positive correlation. In contrast, in London, variables like education density, parking lot density, economic vitality, and Catering density exhibited strong spatial autocorrelation, with Moran’s \(\:I\) values of 0.967, 0.955, 0.952, and 0.941 respectively.

Table 3 Results of Global Moran’s I.
Full size table

Local Indicators of Spatial Association (LISA) cluster maps were used to depict clustering phenomena of hotspots and coldspots and display the local Moran’s \(\:I\) values related to these clusters. Using GeoDa for calculating Moran’s\(\:\:I\), a spatial autocorrelation test was conducted on the visitor emotion data in urban parks of Shanghai and London. As shown in Fig. 6, in Shanghai, the visitor emotion hotspots (high-high clusters) are relatively concentrated around the city center where parks are densely distributed, transportation is convenient, surrounding infrastructure is well-developed, and the proximity to urban rivers tends to induce positive emotions in visitors. The coldspots of emotion values (low-low clusters) mainly occur in the southern and northern parts of the study area. Compared to London, the hotspots of visitor emotions in parks (high-high clusters) are primarily concentrated in the central and southern parts of Inner London, interspersed with coldspots (low-low clusters) of emotion values. Additionally, the Moran’s I statistics for visitor emotions in parks in Shanghai and London (with 999 permutations under a geographical distance spatial weighting matrix) were 0.727 and 0.713, respectively, both with significance levels at 0.001 (<0.05). This demonstrates a strong and positive spatial dependency and clustering of emotion values in parks among visitors in both cities.

Fig. 6
figure 6

Local Moran’s\(\:\:I\): (a) Shanghai (b) London (It was generated by GeoDa 1.14.0 software (https://geodacenter.github.io/)).

Full size image

MGWR results

The analysis of diagnostic indicators (Table 4) suggests that the MGWR (Multiscale Geographically Weighted Regression) model reveals significant spatial non-stationarity in the relationship between visitors’ emotions in urban parks and predictive variables. Through the MGWR results in Table A3, we observe that the model’s varying bandwidth provides a more detailed and comprehensive perspective for analyzing spatial data relationships. The variables in both Shanghai and London exhibited variations at local or global scales, with all local variables ranging from negative to positive. This variation indicates that the factors influencing visitors’ emotions in parks differ depending on the ___location.

Table 4 Diagnostic information of MGWR results in Shanghai and London.
Full size table

Scale effects

In Shanghai, the bandwidth proportions for distance to subway stations, economic vitality, NDWI, parking lot density, dining density, accessibility, and road intersection density range between 1.54% and 7.9%. This indicates that these variables mostly affect visitors’ emotions on a local scale and have significantly different impacts at different locations. Conversely, the effect of NDVI is presented on a larger scale, with a bandwidth distance of 2140, accounting for 76.89% of the total sample size, indicating relatively weak spatial non-stationarity. In contrast, the influence scales of variables such as distance to bus stations, shopping centres density, leisure density, and education density are 2783 (a global scale), suggesting almost no spatial heterogeneity in these variables.

Compared to Shanghai, London shows smaller bandwidths for subway station distance, shopping centres density, leisure density, and parking lot density, reflecting their significant local-scale impacts. The bandwidths for accessibility and bus station distance are 2312 and 2364, respectively, showing lower spatial heterogeneity. The spatial scales for road intersection density, NDVI, economic vitality, NDWI, educational and cultural density, and catering density are 3801, indicating that these global variables do not vary significantly across space, suggesting that the effects of these factors are consistent across different locations.

Fig. 7
figure 7

Spatial distribution of factors local coefficients in Shanghai and London (It was generated by ArcMap 10.6 software (https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview)).

Full size image

Geographic relationship between park environmental factors and visitors’ emotions

In Shanghai (Fig. 7(a)), as shown by the results in Table A3, economic vitality and accessibility are statistically significant variables, with differing directions and magnitudes of influence on visitors’ emotions. Notably, we found a positive correlation between economic vitality and visitor emotions, a finding supported by previous studies43. Urban areas with higher economic vitality typically host more activities, production, and service industries, while also benefiting from greater government fiscal resources for the construction and maintenance of public spaces44. As a result, urban park facilities in these areas tend to be more well-developed and secure, enhancing the overall attractiveness of parks and providing visitors with a more comfortable and enjoyable experience. Figure 7(b) further illustrates that Yangpu District and Pudong New Area exert the most significant positive impact on visitor emotional well-being values due to their strong economic capacity and abundant urban resources. Notable features such as the China (Shanghai) Pilot Free Trade Zone and the Oriental Pearl Tower, along with the city’s longest premium riverside area, contribute to the enhanced visitor experience in these districts. In addition, we found that accessibility also positively influences the emotional well-being values of urban park visitors7, with a relatively small bandwidth, indicating its impact on visitor emotions at a local scale. It is noteworthy that this finding sharply contrasts with the situation in London. Figure 7(c) shows that in districts such as Minhang, Baoshan, Pudong South, and Hongkou, higher walkability to urban parks is associated with stronger emotional well-being values among visitors.

In London (Fig. 7(d)), Contrary to the results in Shanghai, higher accessibility does not correlate with higher visitor sentiment. As shown in Fig. 7(e), the influence of accessibility intensifies from north to south, reaching its strongest negative impact in areas south of Lewisham and Greenwich. These areas contain a significant amount of industrial land and transportation infrastructure, which may lead to noise disturbances and congestion45, while also lacking diverse public facilities and services. As a result, despite the presence of large urban parks, the poor quality of pedestrian infrastructure may increase negative emotions among visitors during their walk. Contrary to existing research findings46, an increase in parking lot density appears to trigger more negative emotions, and the spatial distribution of these variables exhibits significant heterogeneity. Specifically, this factor has a relatively minor negative impact in the eastern boroughs of Tower Hamlets, Greenwich, and Lewisham, while showing localized positive effects in the western boroughs of Wandsworth, Hammersmith & Fulham (H&F), and Westminster, as illustrated in Fig. 7(f). This discrepancy may be attributed to the insufficient parking supply in these areas, caused by dense urban spaces and high land-use intensity, which negatively affects sentiment. Additionally, Fig. 7(g) shows a positive correlation between the Education density and visitor emotional well-being values, with its influence increasing gradually from east to west. Areas with a high concentration of such facilities often receive greater government investment and benefit from higher levels of urban planning, including better park maintenance, more aesthetically pleasing landscape design, and safer environments. These factors enhance the overall attractiveness of parks and improve visitor satisfaction, making it easier for visitors to experience positive emotions. Finally, our results indicate that shopping density (Fig. 7(h)) has a bandwidth of 45, reflecting significant spatial heterogeneity. Its spatial distribution pattern is almost the opposite of parking density, with a strong positive effect observed in eastern areas such as Greenwich, Newham, and northern Lambeth.

Discussion

Application of online review data and NLP methods in visitor emotions

This study aims to explore the positive emotions of park visitors through quantitative analysis of social media data, comparing potential connections between parks and visitor emotions in Shanghai and London. We selected the most influential social platforms in the two regions—Weibo for Shanghai and Twitter for London—as the primary sources for text data. The results indicate that both Weibo and Twitter are equally effective in capturing visitors’ positive emotions, demonstrating the similarity and comparability of social media data across different cultural contexts, despite potential sampling biases arising from differences in data collection methods. The textual data were analyzed using NLP techniques for sentiment analysis. Compared to traditional lexicon-based methods, NLP enables a deeper parsing of sentence structures, accurately capturing emotional tendencies and facilitating large-scale automated analysis, thus providing an advanced solution for cross-cultural emotion research.

Relationship between urban parks and visitor emotions in Shanghai and London

This study employed machine learning techniques to predict relevant variables and constructed an MGWR model on a nonlinear regression basis, uncovering some new insights. The results show that in Shanghai, visitor emotions are most positive in medium-sized parks in the northern central city area. This phenomenon is closely linked to the local economic conditions and the level of public services provided by the government44. On one hand, the northern central area of Shanghai has a higher level of economic development and residents generally have higher income levels47, which translates to a stronger demand and greater spending power for high-quality park green spaces. On one other hand, these areas receive more resources and policy support from the government, creating a favorable public green space environment that significantly enhances visitors’ positive emotions. Furthermore, the diversity of facilities and rich activity options in medium-sized parks also bring a varied experience to visitors13,30. Notably, economic vitality and accessibility both exhibit a significant positive correlation with visitors’ positive emotions in urban parks. Areas with higher economic vitality often host more cultural events, such as music festivals and holiday celebrations44, which are frequently held within or near parks. These events provide diverse recreational opportunities for visitors, enhancing their emotional experiences. Additionally, economically vibrant areas tend to place greater emphasis on the beautification of public spaces, resulting in more refined park designs and landscapes that serve as important venues for social interaction and engagement9. Furthermore, a well-maintained park environment helps visitors relieve stress and improve their overall emotional well-being. Additionally, consistent with previous research, high walkability is positively correlated with visitor emotional well-being values. According to Chuang7, the distance to parks and their accessibility are key factors affecting positive emotions, where improved accessibility significantly enhances the park visiting experience and its utilization rate. Areas such as Minhang, Baoshan, and southern Pudong are predominantly residential and located farther from the city centre, with relatively dispersed commercial and entertainment facilities48. In these areas, higher walkability effectively reduces travel-related stress, increases park visitation frequency and visitor satisfaction, and ultimately enhances the overall visitor experience. Furthermore, these districts often contain medium-sized parks, which, due to their distance from the city centre, tend to feature better greenery and a more tranquil natural environment. These parks provide accessible natural spaces where visitors can enjoy landscapes such as trees, lawns, and lakes. Direct interaction with nature has been shown to relieve stress and have significant positive effects on mental well-being3.

The results of the study in London exhibited different results from Shanghai. In London, the small-sized parks in the southern part of Inner London exhibit the highest positive emotions among visitors. This is likely because the southern area has a long history and many traditional English garden-style small parks attract numerous visitors. Recent large-scale urban renewal efforts, particularly the improvement and enhancement of park green spaces, have further optimized visitors’ emotional experiences49. However, high accessibility does not necessarily enhance visitors’ emotional well-being values, which stands in stark contrast to the findings in Shanghai. A possible explanation is that in London, high walkability does not always equate to a comfortable walking environment. For instance, in the southern parts of Lewisham and Greenwich, despite their high walkability, these areas have relatively lower socioeconomic status and may face issues such as safety concerns and poor environmental hygiene45. These factors likely diminish visitors’ sense of security and comfort, thereby negatively impacting their positive emotions toward the parks. Additionally, although large and medium-sized parks are present in these areas, their landscape features may be less visually appealing compared to parks in other districts. Many of these parks have older planning and design and lack adequate recreational facilities and interactive spaces, making it difficult to enhance visitors’ emotional well-being values through visual and experiential engagement. Previous studies suggest that London visitors tend to prioritise a park’s visual appeal and overall image (Imageability) when choosing a park to visit16. If it underperforms in these aspects, accessibility alone is insufficient to attract visitors. Interestingly, education density is positively correlated with visitors emotional well-being values, as research suggests that environments rich in cultural and educational resources can stimulate creativity, foster a sense of belonging, and enhance overall satisfaction24. These resources provide intellectual and artistic enrichment and contribute to a culturally vibrant atmosphere around urban parks. For example, the British Museum and University of London, both located in northwestern Inner London, are situated near parks, attracting a highly educated population that tends to be more health-conscious and places greater emphasis on well-being and psychological experiences15, thereby influencing the overall emotional state of park visitors. Moreover, urban parks and cultural-educational institutions can create a synergistic effect, as seen in London’s Royal Parks, such as Regent’s Park and Kensington Gardens, where visitors may attend art exhibitions or academic lectures before relaxing in the park, fostering a seamless cultural-recreational experience that enhances positive emotions. In contrast to previous studies7, we also found that high parking lot density does not contribute to visitors’ positive emotions, which may reflect the indirect negative impact of high land-use intensity and urban congestion, reducing visitors’ emotional well-being in parks. Finally, our findings indicate that shopping centres density also has a positive effect on visitor emotions, as greater retail activity not only provides entertainment and a sense of achievement but also expands the functional role of parks, increasing visitors’ time spent in and around these spaces and enhancing their overall positive experiences.

Cross-cultural differences in park planning recommendations

The in-depth exploration of cross-cultural comparisons is crucial for revealing insights into international planning and design principles across different socio-cultural contexts. Specifically for urban park renovation and planning, adopting culturally and regionally differentiated strategies not only promotes positive emotions among people but is also key to enhancing human-centered park planning. Taking Shanghai and London as examples, planners might consider the following aspects:

For Shanghai, government authorities should optimize the allocation of urban park resources. Specifically, it is recommended to enhance park infrastructure by adding sports facilities, nighttime lighting, and water landscapes while also encouraging government or corporate-organized cultural events, such as music festivals, art exhibitions, and fitness classes, to strengthen parks’ social and cultural attributes. Additionally, improving the pedestrian and slow-traffic systems around parks, enhancing accessibility at park entrances, and expanding service coverage areas would further increase visitor engagement. However, the impact of parks on emotional well-being is not solely dependent on walking distance. A more effective strategy is to enhance the parks’ visual appeal and overall environmental aesthetics. For instance, introducing diverse plant species, themed gardens, and ecological water features can significantly enrich visitors’ visual experiences, help alleviate physical and mental fatigue, and promote stress relief.

For London, improving the quality of the walking environment should take priority over merely enhancing accessibility. Green space planners should focus on improving the quality of routes leading to parks, including well-designed sidewalks and safe pathways, as these are key to enhancing the overall park visit experience. Additionally, safety concerns must not be overlooked. Measures such as upgrading park lighting systems, optimising surveillance and emergency response facilities, and raising public safety awareness should be implemented to establish a comprehensive security system, ensuring that visitors can enjoy a safe and comfortable green space. Furthermore, fostering collaborations between parks and nearby museums, universities, and theatres can enrich the visitor experience. Initiatives such as joint ticketing programmes, academic lectures, and outdoor art exhibitions can transform urban parks into integrated spaces for both education and leisure, attracting visitors with a strong interest in cultural and educational activities.

Limitations

This study has several limitations. First, it primarily relied on text data; future research could incorporate various data types, such as video and image data, to more accurately assess visitor sentiment. Second, the study’s sample was predominantly younger social media users, which may not represent all age groups. Third, inconsistencies in the methods used to collect Weibo and Twitter data may have led to discrepancies in the results. Future studies should aim to standardize data collection methods to reduce potential biases. Finally, cross-cultural differences may be influenced not only by the parks themselves but also by variations in visitors’ social behaviors, lifestyles, or social backgrounds28. For example, China’s highly competitive lifestyle and London’s more relaxed social environment may differently shape park visitors’ emotions and preferences. Therefore, future research could benefit from integrating interdisciplinary perspectives (e.g., sociology, economics) rather than focusing solely on the physical attributes of parks.

Conclusions

Urban parks and green spaces are commonly believed to contribute to visitors’ positive emotions and mental health, yet the link between the specific environments and features of urban parks and visitor emotions has been understudied. Using Shanghai and London as cases, this study introduces a novel and comprehensive analytical framework using social media data. We employ machine learning models and spatial autocorrelation to address issues of spatial heterogeneity, and use the MGWR model to provide geographically localized explanations for our findings. This approach allows us to identify nonlinear relationships between different environmental factors of urban parks and positive visitor emotions across different cultural contexts. This research makes three key contributions to park planning:

First, by analyzing the link between positive emotions exhibited by park visitors and park environments across different cultural contexts, this study identifies key factors influencing positive visitor emotions. The study found that in Shanghai, medium-sized parks with high economic vitality and accessibility significantly enhanced visitors’ emotional well-being values. In contrast, in London, high accessibility alone was insufficient to stimulate visitors’ emotional well-being values. Instead, small parks with high education density and low parking lot density were also effective in improving visitors emotional well-being values. These findings offer a new perspective on how urban parks influence visitors’ emotional well-being states.

Second, the study reveals commonalities and differences in visitor emotions and preferences across different cultural contexts. By identifying these similarities and disparities, the research deepens our understanding of urban parks in a cross-cultural context. This process not only enhances our comprehension of the demands and values of urban parks under different cultural backgrounds but also guides the development of more precise park planning and design strategies tailored to diverse cultural groups, ensuring that local needs are effectively met.

Lastly, the study presents a transferable workflow utilizing online reviews and natural language processing (NLP) technologies to quantify urban visitor emotions. This data-driven approach provides urban planners and decision-makers with a powerful tool for more accurate and comprehensive decision support, facilitating more responsive and participatory landscape planning and management.