Introduction

More than 55% of the global population now resides in urban areas1, and the urban population is rapidly increasing. To accommodate the growing population, urban areas must expand or densify. However, this urban expansion or urbanization often comes at the expense of green spaces, particularly in developing countries2,3. Parks constitute a vital element of urban landscapes. Beyond delivering essential ecosystem services, such as temperature regulation and stormwater management, Parks serve as pivotal spaces for public leisure and social interaction. They play an indispensable role in enhancing urban environmental quality and substantially elevating the overall well-being of urban residents4,5.

An increasing body of literature suggests that access to parks contributes positively to the mental health of urban populations6 and can lower the likelihood of developing certain mental health conditions across the general populace7. In urban areas, residents benefit from both natural and cultural ecosystem services provided by proximate parks, facilitated through diverse mechanisms. According to Attention Restoration Theory8, natural environments meet the psychological needs of the public. Additionally, physical activity9, social interaction10, and environmental regulation11 may also mediate psychological benefits.

NDVI and GVI are used to capture public perceptions of green spaces12,13. NDVI is a vegetation index based on satellite imagery, used to quantify urban greening14. Typically, after preprocessing remote sensing images, NDVI is calculated using the reflectance data from the red (Band 4 RED) and near-infrared (Band 5 NIR) bands15. GVI refers to the amount of greenery visually perceptible from a specific ___location. It is calculated based on semantic segmentation of street view images, determining the proportion of vegetation pixels relative to the total pixels in each image to derive the GVI value16.

Social media data (SMD), with its autonomy and spatiotemporal characteristics, is increasingly used to study public perceptions and emotional preferences towards urban spaces17,18. It offers valuable insights into individuals’ interactions with public spaces by capturing pertinent environmental factors19. In contrast to other data sources like street view, sensor data, and traffic information, SMD stands out for its diversity, subjectivity, and spontaneity, offering a direct representation of people’s preferences and responses20,21. Research based on SMD in urban parks has confirmed that public emotions vary in different environments, with individuals exhibiting higher levels of happiness in natural landscapes compared to densely urbanized areas22. For instance, studies have shown that parks outside Manhattan in New York City elicit more positive emotions than those within the city23; compared to neighborhood parks, visitors display more positive emotions towards comprehensive and heritage parks24. These studies validate the effectiveness of SMD in exploring public emotions and urban parks.

There are numerous methodologies for researching the efficacy of parks on public emotion. Multiple linear regression (MLR) is commonly used. For example, Miao et al. employed MLR to analyze the health effects of parks on the elderly. The results indicated that the proximity to parks, park size, and NDVI demonstrated trade-off effects on sedentary behavior, while GVI did not25. However, the MLR model assumes a linear relationship between independent and dependent variables and neglects the influence of spatial ___location. To address this, some scholars use spatial regression models to clarify the spatial attributes between variables. For instance, Cao et al. adopted the geographically weighted regression (GWR) model to examine the spatial correlations between public emotion and park characteristics from three dimensions12. Additionally, when data exist across different hierarchical levels and groups, Hierarchical Linear Modeling (HLM) is often utilized26. HLM, an extension of OLS regression, adapts the traditional OLS method to handle multi-level data structures27. For example, Xie et al. used HLM to explore the relationship between soundscapes and visual environmental elements, considering the impact of differences among 17 historical towns28; Chen et al. evaluated the relationship between overall and individual park quality and social interactions using HLM27.

Existing studies have used SMD to explore the impact of urban green spaces and parks on public emotions, focusing on the use of objective indicators (such as size, type, and vegetation cover) to assess the quality and characteristics of green spaces and parks, and how these indicators influence public emotions and health23,29,30. For example, Aaron J. Schwartz’s research used Twitter data to examine how different types of outdoor public spaces affect subjective well-being. The results found that emotional well-being significantly increased during park visits and remained elevated for several hours after the visit. Regional parks and neighborhood parks were found to be more restorative than civic plazas, providing mental health benefits to park visitors31. Plunz et al. used geo-tagged Twitter data to reveal the correlation between residents’ well-being and urban green spaces. The study found that emotions expressed in tweets varied by area inside and outside the park. In Manhattan, tweets from within parks expressed fewer positive emotions compared to tweets outside the parks, but park visitors in New York City experienced more positive emotions23. Furthermore, the subjectivity, spatiotemporal characteristics, and the relationship between SMD and environmental factors have become a focus of recent research. Liang et al. used multidimensional SMD to assess the visitation of 300 parks in Shanghai, quantifying the number and intensity of visits, revealing the uneven spatiotemporal distribution of park visits32. The results showed that parks in the city center attracted more visitors, and people preferred to visit parks in spring and on rest days. In summary, there is a growing body of research using SMD to study the impact of urban green spaces (UGSs) and parks on emotional perception. However, these studies rarely consider combining “INPY” as a dynamic factor with traditional green space park evaluation indicators, and lack attention to the specific dimension.

With the development of society, public perception of urban parks reflects, to some extent, the popularity of these parks on the internet, leading to the emergence of parks with high popularity and online attention. Compared to other green spaces, high-popularity parks often play a more significant role in people’s daily lives33. However, existing studies rarely consider the “INPY” feature of urban parks, and overlooking this characteristic may introduce bias into the research results. Therefore, when studying the impact of parks on public emotions, in addition to considering traditional features such as park grade and size, INPY should be included as a new influencing factor in the emotional evaluation framework. However, research integrating INPY of urban parks into public emotional analysis remains scarce. Moreover, existing studies have paid little attention to the impact of seasonality on public perception. Seasonality, as an important factor influencing various aspects of urban parks, can significantly alter public perception and emotional responses. Ignoring seasonal changes may lead to misunderstandings regarding the relationship between public emotions and urban parks. Therefore, considering the impact of seasonality on urban parks, especially in the context of its interaction with INPY, will better reveal the multidimensional influence of parks on public emotions under different environmental conditions.

Therefore, this study, based on Weibo check-in data and combined with data from various sources such as street views and remote sensing, investigates the relationship between the INPY of urban parks and public emotions. To refine the differences in ETPN, public emotions are categorized into five gradients, arranged from low to high as follows: N (Negative), RN (Relatively Negative), M (Moderate), RP (Relatively Positive), and P (Positive). Under varying scales of variables, MLR, Multitask regression, and HLM are employed to quantify the association between parks and public emotions. Additionally, to capture the spatial factors influencing emotional differences, text clustering is used to analyze spatial environmental factors. We address the following three questions: (1) Are the INPY of urban parks, GVI, and NDVI related to public emotions? (2) Do different seasons affect the INPY and public emotions? Are there differences between emotional dimensions? (3) Which environmental factors have the greatest impact on public emotions?

Materials and methods

Study area

Luoyang, located in the western part of Henan Province, China, lies in a transition zone between the second and third-level terrain of the country, characterized by complex and varied landforms. It boasts abundant natural resources and profound historical and cultural heritage. The total area of Luoyang’s central urban district spans 15,230 square kilometers, with the urban area covering 2,274 square kilometers. The city has a permanent population of approximately 7.05 million. Luoyang is a major metropolis with advanced industries, commerce, tourism, and more. The central urban district typically includes the Old City, Xigong, Jianxi, Luolong, Chanhe Hui, and Yanshi District. According to the “2020 Henan Livable Cities Ranking,” Luoyang ranks among the top three cities in terms of livability(https://www.henan.gov.cn/). Therefore, focusing on the parks and green spaces in the central urban district of Luoyang offers both typical relevance and practical significance for this study. Through the Weibo check-in platform, using “urban parks” as the search keyword, and supplementing with online map platforms such as Baidu Maps for further filtering, we ultimately selected 50 parks as the research subjects (Fig. 1).

Fig. 1
figure 1

Study area.

Full size image

To better study the impact of park characteristics on public emotions, the park types, grades, and sizes were determined based on relevant literature34,35,36 and the “Urban Green Space Classification Standards” (Table 1, for detailed classification results see Supplementary Table S1).

Table 1 Park grades, types, and sizes.
Full size table

Research framework

The experimental design and process are shown in Fig. 2, with parks in the central urban area of Luoyang selected as the research subjects. This study is based on Weibo check-in data, supplemented with data from multiple sources such as street views and remote sensing, to investigate the relationship between INPY of urban parks and public sentiment. First, the SnowNLP method was used to analyze Sina Weibo text data to calculate public sentiment perception, categorizing sentiment into five gradients: N, RN, M, RP, and P. The INPY was calculated based on Weibo check-in frequency, likes, reposts, and comments. Subsequently, green space exposure at different scales (e.g., NDVI, GVI) around the parks was collected as objective environmental indicators, while also considering the seasonal effects on each variable. A regression model was constructed using ETPN and ETGD as dependent variables, and INPY, NDVI, and GVI as independent variables. The study employed correlation analysis, MLR, Multi-task regression, and HLM to quantify the relationship between parks and public sentiment. Additionally, to capture spatial factors influencing emotional differences, text clustering was used to analyze spatial environmental factors, aiming to identify urban park environmental features that influence visitors’ positive emotions (Fig. 2). This research framework integrates the dynamic factor of “INPY” with traditional park evaluation indicators, considering the potential impact of this emerging factor on sentiment perception, filling a gap in related research.

Fig. 2
figure 2

Research framework.

Full size image

Research data

Weibo is one of the major social media platforms in China37. 60.8% of new media users consider platforms such as Weibo as their primary source of information over the past three months. With its rapid dissemination, strong interactivity, and massive scale, Weibo has become a major platform for users to obtain information in the internet age (https://mall.sc.weibo.com/datacenter/). Therefore, using it as a data source for studying the INPY and public emotions is both representative and of practical significance.

Data acquisition

In this study, we collected a year’s worth of Weibo data from 2023 for Luoyang, representing one of China’s major social media platforms. The data collection was performed using Python scripts, with key fields selected including user ID, text content, posting time, latitude, and longitude. The raw data was stored in JSON format and ultimately converted into CSV format. During the data cleaning process, we excluded posts that were outside the geographic scope, duplicates, non-Chinese content, blank expressions, and those with imprecise geographic locations38,39. We also filtered out content likely posted by bots, such as advertisements and weather reports22. Additionally, to study the relationship between public emotion and the environment, we excluded data not covered by Baidu Street View. In the end, we collected a total of 222,810 geotagged Weibo posts, of which 41,675 were from 50 urban parks of different grades and sizes. The Weibo check-in frequency, number of reposts, comments, and likes from these 50 parks were used to calculate INPY, while the Weibo texts were used to calculate ETPN and ETGD.

Given the concerns surrounding data privacy and the ethical implications of using social media data, all information was sourced from publicly available posts by Sina Weibo users. In compliance with privacy regulations, no personal details such as age or gender were collected. The data was also anonymized to ensure confidentiality. This paper presents only aggregated results, ensuring that individual users cannot be identified from the reported information.

Data processing

For large-scale urban parks such as the Longmen Grottoes, there may be multiple check-in points within a single area, with significant differences in popularity between these points, which could affect subsequent analysis. Based on this, the study handled such data in two ways.

Firstly, if there is a large frequency difference between the primary and secondary check-in points, the data from the less frequented points are directly excluded. For instance, in the Longmen Grottoes scenic area, where the main check-in point accounts for 97% of the data, only the data from this major category is retained, representing the overall situation.

Secondly, if the frequency difference between the primary and secondary check-in points is small, the calculation is based on the proportion of Weibo posts40. For example, in the case of Wan’an Mountain Scenic Area and Wan’an Mountain Summit Park, the frequency ratio serves as the weight for each, culminating in aggregated data for the Wan’an Mountain Scenic Area.

INPY

The INPY of various parks is not solely dependent on check-in frequency but also considers the combined effects of the number of reposts, comments, and likes. This study adapts a formula used by TikTok to calculate popularity scores (i.e., Popularity Score = Check-in frequency * a + Number of reposts * b + Number of comments * c + Number of likes * d)41,42. Using the Analytic Hierarchy Process (AHP) combined with expert scoring, the weights for each component were determined as follows:

$$H = F \times 0.5688+R \times 0.0659+C \times 0.128+L \times 0.2372,$$
(1)

H represents INPY; F is the Check-in Frequency; R is the Number of reposts; C is the Number of comments; L is the Number of likes.

AHP is a structured method for analyzing complex problems, relying on expert judgment to determine priorities43. It is widely used for assigning weights in decision-making44. While subjective, these weights are based on expert insights and are highly relevant and practical in the current context. Using the above formula, INPY for 50 parks for the entire year of 2023 and for each of the four seasons was calculated (Fig. 3).

Fig. 3
figure 3

Spatial Distribution of INPY: (a) Annual; (b) First Quarter; (c) Second Quarter; (d) Third Quarter; (e) Fourth Quarter.

Full size image

ETPN and ETGD

The sentiment analysis of comment texts was performed using the SnowNLP library in Python. SnowNLP, a suite of tools based on Bayesian models, is specifically designed for processing Chinese natural language and conducting sentiment analysis (https://github.com/isnowfy/snownlp). The formula for sentiment analysis is as follows:

$$P({c_i}|{w_1}, \cdots ,{w_n})=\frac{{[P({w_1}, \cdots ,{w_n}|{c_i}) \times P({c_i})]}}{{[P({w_1}, \cdots ,{w_n}|{c_1}) \times P({c_1})+P({w_1}, \cdots ,{w_n}|{c_2}) \times P({c_2})]}},i=1,2$$
(2)

Among them, \({c_1}\)and \({c_2}\) respectively represent the positive and negative emotions of the sample, \({w_n}\) represents the occurrence probability of the specific emotional feature word w in the test sample, \(P({c_i})\)represents the probability that sentences containing the feature word \({w_n}\) belong to the emotional category \({c_i}\).

To enhance the accuracy of SnowNLP in sentiment analysis, 120,000 valid data points from other regions were collected using Python tools and manually labeled as positive and negative emotions. These categorized data were split into training, validation, and test sets with a ratio of 7:1.5:1.5. The model’s performance was evaluated using various metrics such as cross-validation, confusion matrices, accuracy, and recall to ensure its accuracy and reliability. The final trained model achieved an accuracy of 0.97 and a recall rate of 0.96, effectively reflecting public emotional tendencies (Supplementary Fig. S1 and S2). To minimize the impact of irrelevant terms on the analysis, stopwords were incorporated. The evaluation text was further refined by integrating the Jieba tokenizer for improved segmentation. The output values of sentiment analysis range from 0 to 1, where values close to 1 indicate more positive emotions in the comments, and values close to 0 indicate more negative emotions (Table 2).

To further refine the perception differences among the public, we have established five categories for the ETGD: 0-0.2 as N, 0.2–0.4 as RN, 0.4–0.6 as M, 0.6–0.8 as RP, and 0.8-1 as P. Detailed classification results can be found in Supplementary Table S2 and S3.

Table 2 Examples of sentiment analysis in evaluation texts.
Full size table

NDVI and GVI

Green space exposure typically includes NDVI and GVI, which are used to measure the quantity and visibility of green space exposure45.

Satellite images used for calculating NDVI were obtained from the United States Geological Survey (USGS) official website (https://earthexplorer.usgs.gov/). Initially, cloud-free images from the Landsat 8 OLI_TIRS, taken on June 23, 2023, with a resolution of 30 m, were selected. These images underwent geometric and atmospheric corrections using the Semi-Automatic plugin on the QGIS platform. The NDVI values range from − 1 to + 1, with vegetated areas typically showing positive values (Fig. 4). Negative values, usually indicative of water bodies and beyond the scope of our study, were excluded from the calculations near parks46.

Street view images for calculating the GVI are obtained from Baidu Online Maps, where green vegetation pixels are extracted to determine the GVI. Road network data provided by OpenStreetMap (OSM) are used to set a sampling point every 50 m. The coordinates for these points are inputted into a Python script utilizing an Application Programming Interface (API). For each point, images from four directions (0 degrees, 90 degrees, 180 degrees, and 270 degrees) are captured38, totaling 175,284 images.

To analyze these images, this study employs the urban landscape dataset Cityscapes (https://gitcode.com/ghmirrors/cityscapesScripts). Utilizing the PyTorch deep learning framework and the pre-trained DeepLabV3Plus-MobileNet model allows for rapid visualization processing of images and their pixel-level labels (https://github.com/VainF/DeepLabV3Plus-Pytorch). The existing dataset can identify 30 types of street view elements, including various vertical (such as trees and hedges) and horizontal vegetation (such as grass and soil). GVI is quantified by calculating the sum of the pixels of vertical and horizontal vegetation relative to the total pixels in each image (Fig. 5).

Fig. 4
figure 4

NDVI.

Full size image
Fig. 5
figure 5

GVI.

Full size image

Research on the impact of distance on residents’ perception of green spaces47 has demonstrated that green spaces within 500-meter and 1,000-meter buffer zones show a strong correlation with public emotions48,49,50,51. This is because distances of 500 m and 1,000 m generally correspond to a 10 to 15-minute walk, which covers the majority of daily activities for most people47. Considering that some parks are too small, a 200-meter buffer zone was also adopted. Ultimately, the average NDVI and GVI within 200 m, 500 m, and 1,000 m circular buffer zones around each park were calculated and used as indicators of green space exposure.

Research methods

Statistical models and framework

In this study, the dependent variables include ETPN and ETGD. For independent variables, the INPY of urban parks is selected as a metric for assessing social attention. Additionally, the study involves NDVI and GVI at different spatial scales (200 m, 500 m, and 1000 m). To address the various data types and research objectives, this study employs multiple regression models including MLR, MultiTask LassoCV, and HLM for analysis.

MLR (Table 3. Model 1) is commonly applied in urban studies due to its strong interpretability52,53,54. To determine the optimal set of variables for MLR, the process begins with Pearson correlation to identify key variables that explain DBS volume, eliminating those with no significant impact. Next, the Variance Inflation Factor (VIF) is calculated to assess potential multicollinearity among the independent variables, removing those with a VIF exceeding 10 53.

Given that ETGD are divided into five intervals: N, RN, M, RP, P, the study employed a MultiTask LassoCV model (Table 3. Model 2) to examine the effects of multiple variables on each emotional interval. This model determines the optimal regularization parameter through built-in cross-validation, enabling effective prediction across multiple output variables. It not only reveals potential associations between INPY, GVI, and NDVI with emotional states but also provides a new perspective through the Multi-task learning framework on how environmental factors collectively influence human emotional responses.

To study the impact of seasons on the relationships among various factors, HLM was introduced26, to capture the variability and influence at each level of the hierarchical structure. Since NDVI and GVI represent data from a specific point in time and do not possess attributes that change over time, in the HLM, NDVI and GVI were incorporated as static variables (variables that do not change over time). These serve as background or control variables to help adjust the model to explain the impact of other more frequently changing variables, such as the INPY of parks, on ETPN. Additionally, seasons were set as random intercepts to analyze the differences in the effects of independent variables (INPY, GVI, and NDVI) on dependent variables (overall emotional value and ETGD) across different seasons (Table 3. Model 3). This approach reveals how seasonal factors modulate emotional dynamics, allowing for a deeper understanding of the mechanisms behind emotional fluctuations. Descriptive statistics for each model can be seen in Supplementary Table S4 and S5.

Table 3 Summary of statistical models.
Full size table

Text clustering

To deeply analyze the factors influencing public emotions in Luoyang city parks, this study conducted topic extraction and text clustering on park comments. The key steps are as follows:

(1) Key phrase extraction.

First, the text was segmented using Jieba and redundant words were removed with a Chinese common stopword dictionary. Then, term frequency-inverse document frequency (TF-IDF) was used to identify key phrases. In this study, the TF-IDF score was calculated for each label. This score is proportional to the frequency of occurrence of the label in specific clusters of users or images and is offset by the total number of users or images associated with that label. For positive and negative texts, the top 50 labels with the highest TF-IDF scores were identified, and the weights of these words were calculated (Supplementary Table S6 and S7).

(2) Keyword vectorization.

The 50 key labels identified were converted into 100-dimensional word vectors using the Tencent AI Lab Embedding Corpus55, which includes 8 million Chinese words and is capable of capturing subtle semantic differences of Chinese words and phrases. In this step, the gensim library in Python was used to load and manipulate the word vector model.

(3) Keyword clustering.

Clustering was conducted with the k-means algorithm from Python’s scikit-learn package, which minimizes the sum of squared distances between data points and their assigned cluster centers to optimize the cluster assignments. The optimal number of clusters was selected based on the average silhouette width. The UMAP algorithm was applied to reduce dimensions to two for visualization purposes, and a scatter plot was used to display the spatial distribution of clusters, with each point’s size reflecting its weight and color distinguishing different clusters.

Results

Relationships between INPY, NDVI, GVI, and public emotion

Relationships between INPY, NDVI, GVI, and ETPN

The coupling of INPY with spatial emotions was statistically analyzed according to geographic coordinates. As shown in Supplementary Fig. S3, the coupling degree between INPY and spatial emotions is not high. We then employed the Spearman correlation coefficient to examine the relationships between INPY, NDVI, GVI, and spatial emotions. The results indicate that the correlations between INPY, different scales of NDVI, GVI, and spatial emotions are not significant (Fig. 6).

Fig. 6
figure 6

Spearman correlation analysis results. *Significant at the 0.05 level. **Significant at the 0.01 level.

Full size image

We employed a MLR to investigate the relationships between INPY, NDVI, GVI, and ETPN. Due to issues of multicollinearity, NDVI_500 and NDVI_1000 were removed from the independent variables. The remaining variables all passed the VIF test (VIF < 10). The results indicate that NDVI at a 200 m scale has a significant positive impact on overall ETPN, suggesting that micro-scale landscape greenery can enhance people’s ETPN. The impacts of other variables were not significant (Table 4).

Table 4 MLR modeling results.
Full size table

Relationships between INPY, NDVI, GVI, and ETGD

Using ETGD as the dependent variable and INPY, NDVI, and GVI as independent variables, correlation analysis was conducted (Table 5). The results reveal a significant positive correlation between ETGD and INPY. Additionally, as the emotional values increase, the correlation coefficient also gradually increases, approaching 1, suggesting a potential linearly increasing relationship between ETGD and INPY.

Table 5 Correlation analysis results for ETGD.
Full size table

We then used Multi-task Lasso regression to further reveal this correlation. The results (Table 6) show that urban parks’ INPY has a significant positive impact on ETGD, particularly on N and M, with regression coefficients of 0.882 and 0.868, respectively. This indicates a close association between the increase in parks’ popularity and positive changes in ETGD. Additionally, the green landscape index at a 200-meter scale shows a negative correlation with N (coefficient approximately − 0.075), suggesting that micro-scale landscape greening can ameliorate negative emotions. Moreover, the impact of NDVI at different scales (NDVI_200, NDVI_500, NDVI_1000) varies across ETGD, particularly the NDVI at 500-meter and 1000-meter scales show a strong positive correlation with P, with coefficients of 0.131 and 0.089 respectively, further confirming the positive correlation between extensive green coverage and citizens’ emotions.

Using the Bootstrap method with 1,000 resamplings, we calculated the 95% confidence intervals of the regression coefficients to strengthen the model’s credibility and scientific rigor. The overall R2 value of the model is 0.613, indicating that the model can explain approximately 61.3% of the variance in ETGD. This reflects the model’s good performance in explaining how urban parks influence citizens’ emotions.

Table 6 Results of multitask LassoCV regression analysis.
Full size table

Relationships between INPY, NDVI, GVI, and public emotion

In this study, to more clearly and visually discern the relationships among variables, we utilized a random forest model to analyze the impact of urban parks’ INPY, GVI, and NDVI on public emotions. These indices are considered key variables for assessing the impact of parks on residents’ psychological well-being56. Additionally, the random forest is widely used in the landscape perception ___domain due to its robust capability to handle high-dimensional data and nonlinear relationships57, such as analyzing the nonlinear relationships between tourists’ positive emotions towards urban parks and influencing factors24.

By constructing six independent models, we revealed the specific impacts of different environmental indices on overall ETPN and various ETGD (N, RN, M, RP, P). The results (Fig. 7) show that INPY has the most significant impact on public emotions. This result seems to contradict the relatively weak correlation between INPY and ETPN discussed previously, mainly because INPY influences spatial emotions through nonlinear pathways or interactions with other variables in the random forest model. As shown in Fig. 7(a), the importance of each factor is relatively balanced. Furthermore, the influence of INPY may be more significant under specific conditions or within certain data subsets, which may not be easily detected in global correlation analysis.

Fig. 7
figure 7

Feature importance of random forest.

Full size image

Relationship between popularity and emotion across different seasons

The HLM analysis results (Table 7) show that, for overall ETPN, INPY has a slight positive effect on ETPN, with a coefficient of 0.001. This differs from the previously discussed results, likely due to interactions across different quarters, where INPY does have an impact on ETPN at the quarterly level. The effects of NDVI and GVI on ETPN also vary: within a 1000 m range, GVI has a significant positive effect on ETPN, while NDVI has a negative effect; within a 500 m range, NDVI has a positive effect on ETPN. The variance estimate of Group Var in the random effects is 0, indicating that the seasonal random effect on ETPN is not significant.

Regarding ETGD, INPY has a significant positive effect on all emotional dimensions (N, RN, M, RP, P), particularly on P, with a coefficient of 0.319, indicating that an increase in INPY strongly promotes positive emotions. Additionally, for RP and P, there is considerable seasonal variation (relatively high group variance), suggesting that seasons may have a stronger impact on positive factors. Environmental and social factors in each season can greatly influence people’s emotional expressions. Furthermore, we found that NDVI at 200 m and 500 m significantly affects N, which aligns with the previously discussed results, indicating a stronger association between micro-scale landscapes and negative emotions.

In conclusion, by constructing a multidimensional analytical framework that includes the “INPY” indicator, we found that INPY is the primary factor influencing emotional expression, with its impact increasing as the emotional gradient rises. The effects of GVI and NDVI vary across different emotional perceptions and gradients. However, large-scale green space exposure is more strongly associated with positive emotions, while small-scale green space exposure tends to have a greater impact on the expression of negative emotions. Additionally, seasonal changes can influence emotional expression to some extent, especially for RP and P.

Table 7 Regression results of HLM.
Full size table

Keyword and clustering extraction results

Keyword extraction results

We extracted keywords from Weibo comments of the top 20 popular parks, classifying them according to their emotional tendency into positive comments (≥ 0.5) and negative comments (< 0.5). The top 20 keywords were extracted for each park based on their weight values (Fig. 8). Detailed extraction results are available in the Supplementary Table S6 and S7.

Overall, the top five keywords in positive comments were “happy,” “joy,” “like,” “hahaha,” and “romantic.” In contrast, the top five keywords in negative comments were “rain,” “too much,” “don’t want,” “subway,” and “pity.” Specifically, for the Longmen Grottoes scenic area, the prominent keywords in positive comments included “joy” (0.032), “magnificent” (0.031), “happy” (0.029), “treasure” (0.012), and “dazzling” (0.012); while in negative comments, “pity” (0.026), “damage” (0.020), “can’t walk” (0.019), “tired” (0.018), “queueing” (0.017), and “crowded” (0.014) had higher proportions.

Fig. 8
figure 8

Keyword word cloud.

Full size image

Clustering extraction results

The optimal number of clusters and their main themes were determined based on the average silhouette score and the interpretability of each cluster. For positive comments (Fig. 9), ten clusters were identified: Visitor Experience, Natural Environment, Traditional Festivals, Service Facilities, Cultural Facilities, Recreational Activities, Transportation Facilities, Entertainment Facilities, Cost, and Park Features. For negative comments (Fig. 10), eight clusters were identified: Ticketing, Service Facilities, Transportation Facilities, Entertainment Facilities, Visitor Experience, Natural Environment, Recreational Activities, and Park Features. Specifically, positive comments primarily consisted of adjectives and verbs, whereas negative comments were predominantly defined by nouns (Table 8). The detailed explanation of each category can be found in Supplementary Table S8.

Fig. 9
figure 9

Clustering of positive comments.

Full size image
Fig. 10
figure 10

Clustering of negative comments.

Full size image
Table 8 Analysis of part-of-speech in positive and negative reviews.
Full size table

Discussion

Variability in the impact of park features on public emotions

Using the SnowNLP algorithm, we obtained emotional scores for each park (Fig. 11). To further explore whether park features influence visitors’ perceptual evaluations, we conducted a correlation analysis between ETPN, INPY, and park features (Supplementary Table S9).

For categorical variables such as park type and grade, we employed the Kruskal-Wallis H test; for continuous variables like park area, we used Spearman’s correlation analysis. The findings revealed no significant correlation between the INPY of parks and their type, grade, or size; larger or higher-grade parks do not necessarily draw more visitors. Similarly, there is no correlation between ETPN and park type, grade, or size, suggesting that the characteristic attributes of parks are not significantly linked to visitors’ recreational experiences.

Fig. 11
figure 11

Distribution of visitor emotions in Luoyang city parks, 2023.

Full size image

This conclusion is consistent with Zou et al.‘s finding that the popularity of parks does not fully reflect visitors’ emotions58. However, discrepancies exist between park type, grade, size, and ETPN, primarily due to regional differences and varying evaluation criteria59. As a renowned tourist city rich in historical and cultural resources, Luoyang’s historical features endow its parks with unique cultural attributes, which differ significantly from parks formed primarily through natural elements60,61,62. Weibo check-in data primarily involves tourist sightseeing and scenic spot check-ins, hence the importance of touristic and sightseeing attributes outweighs that of natural attributes. Ning Zou et al.‘s study focused on urban parks in Chaoyang District of Beijing, which differ significantly from the parks in the central area of Luoyang. Additionally, due to the ambiguity of standards, the determination of park types and grades based on references and relevant norms may lead to significant disparities in results63,64. For example, both the Longmen Grottoes, which ranks first in INPY, and the High Mountain Peony Garden, which ranks last, are categorized as specialized parks of the first grade (Supplementary Table S1).

Differences in the impact of INPY, GVI, and NDVI on public emotions

Integrating Models 1, 2, and 3, along with the random forest model, we observed that a park’s Internet visibility and public interaction levels might directly influence people’s emotional responses and attitudes65, suggesting that enhancing INPY could be a potential strategy for improving public emotions. This indicates the differential impacts of INPY, GVI, and NDVI on public emotions.

Firstly, GVI and NDVI, as measures of vegetation cover and health, exhibit varying impacts on public emotions (ETPN and ETGD) at different scales (200 m, 500 m, 1000 m). For instance, the NDVI score within a 200-meter range accounted for 16.7% in predicting ETPN, demonstrating that closer contact with vegetation more significantly affects public emotions66,67, underscoring the importance of micro-scale greening. While NDVI and GVI have dimensional differences and increasingly divergent impacts on emotions as the scale expands, at smaller scales, their effects on public emotions show similarities47,68.

These results are consistent with earlier studies. In particular, previous research has demonstrated a significant relationship between residents’ well-being and the surrounding greenery69, with increased vegetation cover notably enhancing public emotions70. Similarly, Wang et al. found that greater visibility of vegetation cover correlates with reduced negative emotions71. Building on this, our study further demonstrates that larger-scale GVI and NDVI are more strongly correlated with positive emotions, while smaller scales have a greater impact on negative emotions72. Environmental psychology theories suggest that larger-scale green spaces or natural landscapes generally provide a broader, more tranquil environment, which helps reduce stress and enhance positive emotions. In contrast, smaller-scale environments (such as parks near city streets or specific small areas) may negatively affect emotions due to localized issues like noise pollution and traffic congestion73. Additionally, Bratman et al. also found in their study that the impact of natural environments on mental health varies with spatial scale. Larger green spaces offer more emotional regulation benefits, while smaller-scale environmental changes may more prominently highlight sources of negative emotions74.

Furthermore, the study also found that the INPY of urban parks can similarly affect people’s ETPN, having a significant positive impact on emotions. The degree of this positive impact varies across different ETGD, with a more pronounced effect on positive ETGD. This variation could be attributed to people’s preference for parks that are highly popular Internet, widely shared, and well-known75.

In conclusion, the analysis demonstrates that INPY primarily influences positive emotions, while GVI and NDVI play a role in alleviating negative emotions and enhancing overall ETPN76. These findings align with previous research suggesting that urban parks can elevate public mood23,38,77, validating the potential relationship between the INPY and public perception. This highlights the crucial role of urban parks in enhancing public emotional well-being and provides urban planners with targeted strategies to support residents’ psychological health and emotional well-being.

Seasonal variations in the impact of INPY on ETPN

The INPY of urban parks in the central area of Luoyang varies significantly with the seasons (Supplementary Fig. S4). Specifically, the popularity curve of the Longmen Grottoes exhibits a zigzag pattern, generally trending upward. In contrast, the Luoyi Ancient City saw a continuous decline in popularity during the first three quarters, followed by a gradual increase. The most dramatic changes occurred in the Sui and Tang Luoyang City National Heritage Park, where the popularity curve showed sharp fluctuations, with the most rapid increase in the third quarter.

The reasons for these seasonal variations may be linked to the second quarter coinciding with Luoyang’s traditional Peony Festival, which attracts a surge in domestic and international tourists, with parks featuring Luoyang’s regional characteristics becoming focal points for sightseeing. Additionally, significant spikes in visitor volume occur around traditional Chinese holidays such as Labor Day, National Day, and Chinese New Year, creating substantial public discourse. During these periods, INPY often peaks. For example, the INPY of the Longmen Grottoes scenic area exhibits significant seasonal fluctuations (Supplementary Fig. S5), with two key turning points: the first being the Luoyang Peony Festival in the second quarter, and the second being the Chinese Lunar New Year in the fourth quarter. Additionally, sentiment values also show notable seasonal changes, aligning with the trend of INPY with positive emotions peaking in the second and fourth quarters. These trends indicate that traditional cultural festivals have a significant impact on both INPY and public emotions.

Due to its unique geographical attributes and cultural features, Luoyang has parks with distinct characteristics. These include the Luoyi Ancient City scenic area, known for its historical reminiscence and blend of ancient and modern experiences; the Longmen Grottoes, notable for their grand scale and diverse artistic styles; the Sui and Tang Luoyang City National Heritage Park, rich in cultural heritage and unique architectural styles; and natural scenic parks like Xiyuan Park and Wangcheng Park, which are characterized by their natural scenery and social leisure activities. These diverse types of parks offer different options for tourism and sightseeing across seasons, leading to variations in popularity throughout the year78. As a renowned historical and cultural city, Luoyang’s parks that feature historical relics tend to attract higher popularity. This observation aligns with Yang et al.‘s study on the popularity of tourist attractions in Nanchang, which found that cultural and historical attractions are generally more popular than natural sites79.

The seasonal variations in popularity influence public perceptions of urban parks80, especially in terms of positive comments. The third and fourth quarters are the peak tourism seasons in Luoyang, during which the quality and quantity, depth, and breadth of comments about the parks tend to be more comprehensive. This trend is reflected in the emotional expressions found in Weibo posts, leading to more positive perceptual evaluations80. Furthermore, popular parks, due to their variety of amenities such as sports facilities, recreational facilities, dining services, etc., and diverse activities, can offer visitors a richer experience, thereby enhancing their satisfaction and positive perceptions75. Additionally, parks with high foot traffic usually have more conveniences, such as accessibility to transportation, good infrastructure, and frequent activities, all of which can increase visitors’ frequency of visits and enhance their overall experience. For example, in summer, the shade coverage by vegetation, proximity to water bodies, and the availability of well-maintained facilities significantly increase visitor density and positive perceptions80. Visitors participating in more activities in these popular parks are more likely to achieve psychological restoration and emotional satisfaction78.

Exploring contextual factors influencing public emotions in urban parks

In this study, visitors’ evaluations of the park reflect their needs and expectations for different functional spaces.

Positive reviews often emphasize aesthetics, focusing primarily on the park’s natural landscapes and recreational facilities. Natural features include evaluations of the park’s scenery, such as “spectacular” or “breathtaking,” while recreational facilities encompass performances, food offerings, and fireworks. These elements not only provide visitors with enjoyable visual experiences but also help restore attention, reduce stress, and alleviate cognitive fatigue. Additionally, well-maintained facilities and infrastructure contribute to a more comfortable environment73. Meanwhile, the park’s historical and cultural characteristics strengthen the emotional connection between visitors and the ___location, enhancing their attachment to the park. This, in turn, promotes their appreciation of both the park’s natural landscapes and cultural features, enriching the overall visitor experience81.

However, negative reviews mainly focus on management issues, such as “overcrowding” and “long queues.” These problems may lead visitors to feel that the environment is chaotic and uncomfortable, affecting their psychological state. Social disorganization theory can explain this phenomenon82,83, as poor park management can deteriorate the visitor experience, especially when there are too many people, and a lack of effective management prevents the environment from maintaining order. Strengthening management and maintaining service facilities can not only enhance the overall visitor experience but also help increase the park’s attractiveness.

The differences between positive and negative reviews reflect the varying expectations of visitors regarding urban parks. Urban parks are multifunctional public spaces, and different groups of people prioritize different aspects of their experience. The association between positive emotions and aesthetics may reflect visitors’ high regard for the park’s landscape and environment, as a comfortable, tranquil, and beautiful environment meets users’ expectations for public spaces. These expectations likely drive their positive reviews of the park’s aesthetics84. On the other hand, management issues, particularly problems such as “crowds” and “long queues,” often affect users’ comfort and experience85, which is why negative reviews tend to focus more on management problems.

Additionally, studies indicates that the inherent entertainment quality of parks significantly impacts visitor experience86. In addition to the overall quality of the park, its accessibility and available transportation options play a significant role in shaping visitor assessments27. Therefore, optimizing environmental quality, improving park management practices, enriching activities within the park, providing accessible facilities for vulnerable groups, and maintaining public service facilities are crucial for enhancing the overall visitor experience and increasing the park’s attractiveness87,88,89.

The specific measures to improve the park experience

The results of this study indicate that the INPY, seasonal fluctuations, and management level of parks have a significant impact on public sentiment. Based on this, we propose the following specific park management measures.

Firstly, park managers should utilize emerging technologies and multi-source data analysis to optimize the availability and layout of urban green spaces, with a focus on areas that are highly visible and easily accessible90. According to the views of Hartig, T91. , Giles-Corti, B92. , green spaces with high visibility are often closely associated with positive emotions and psychological restoration. Improving the greening in these areas can significantly enhance visitors’ emotional experiences. For these areas, further enhancing ecological functions through increased vegetation and improved landscape design can elevate positive emotions. Meanwhile, for areas with lower accessibility but higher greening indexes, infrastructure construction should be strengthened, such as adding walking paths and establishing more convenient public transportation links, thereby increasing visitor frequency and alleviating the development of negative emotions.

Secondly, park management should also take seasonal factors into account. Our study shows that the second and fourth quarters coincide with local tourism peak periods, during which the number of park visitors significantly increases, making efficient crowd management particularly important. According to Geng et al.88, seasonal fluctuations affect visitor experiences, especially in high-traffic environments where visitors often feel pressure and dissatisfaction. Therefore, governments and relevant management authorities should align cultural and festive activities with these peak periods, utilizing their characteristics to organize appropriate events and recreational activities that attract visitors and enhance their emotional experience. Additionally, a visitor flow monitoring system and a reservation-based entry mechanism can effectively address overcrowding issues and reduce visitor anxiety and dissatisfaction during peak periods93.

Finally, the government can implement policies that encourage park managers to prioritize the improvement of green areas that both enhance emotional perception and offer high environmental value, especially areas with higher GVI. According to Maller et al.94, improving the quality and accessibility of parks not only enhances visitors’ emotional experiences but also promotes the overall health and well-being of urban residents. Additionally, the government and relevant organizations should adopt a cross-departmental collaboration approach, working together with public transportation, urban planning, and cultural departments to develop management strategies for peak visitor periods. These strategies should include, but not be limited to, adding transportation routes, optimizing walking paths inside and outside the park, and offering convenient cultural and recreational activities during festivals95. By properly planning the park’s spatial layout and time management, park management issues can be effectively reduced, enhancing the park’s social value and cultural function, thereby allowing the park to better serve the public.

Limitations and outlook

While our research has confirmed the diverse impacts of urban parks’ distribution characteristics and Internet popularity on public emotions, there are limitations to our study. First, the emotions expressed by Weibo users at geographic check-in points may not accurately reflect the sentiments of the entire population in those areas, but rather those of a specific subset of individuals38. The use of the AHP method to determine the weights of various indicators in Weibo data involves a certain degree of subjectivity. To enhance the scientific basis of the weight selection, we will further consider conducting a sensitivity analysis to explore the impact of different weight configurations on the results. Third, in interpreting model results, GVI and NDVI represent data points from a specific moment, and the lack of seasonal variation may not fully reflect changes throughout the study period. We treated these as fixed effects to mitigate data short-comings.

Future research should integrate various types and sources of data, such as those from on-site ob-servations, surveys96, environmental psychology97, mobile signaling98, and Geographic Information Systems99. Additionally, other emerging data sources, such as Douyin and TikTok, should be incorporated to complement Weibo data, in order to capture a broader and more diverse range of public sentiment. Furthermore, it should include NDVI and GVI data from different quarters to explore the impact of green space exposure indicators on urban park emotional perception at different times. Considering the role of cultural factors in shaping public emotions of green spaces, these should be incorporated into future models to analyze the interactions between cultural factors and other spatial characteristics in greater depth. Additionally, cross-cultural comparative studies should be conducted to validate the impact of these factors across different seasons and cultural contexts. At the same time, more stratified analyses and advanced statistical methods should be employed to further explore these dynamics, providing a better understanding of the relationship between INPY and spatial emotions.

Conclusion

This study, based on SMD, explores the relationship between the INPY and public emotions of parks. The study found:

  1. (1)

    INPY exerts a positive influence across ETGD (N, RN, M, RP, P) from low to high degrees.

  2. (2)

    GVI and NDVI show considerable variability at different scales and across emotional categories: GVI is negatively correlated with N at the 200-meter scale and has a significant positive effect on ETPN at the 1000-meter scale; NDVI shows a strong positive correlation with P at both the 500-meter and 1000-meter scales.

  3. (3)

    Seasonal variations can influence emotional expressions to a certain extent, particularly with RP and P. Additionally, considering seasonal conditions, INPY has a slight positive influence on ETPN.

  4. (4)

    INPY primarily impacts positive emotions, while GVI and NDVI play roles in alleviating negative emotions and enhancing overall ETPN.

In summary, this research makes significant contributions to the understanding of public perception of urban parks, their INPY, and their characteristic attributes. Firstly, it enriches the methodology for studying urban parks by using Weibo check-in data to capture the public’s intuitive feelings about urban spaces. Combined with satellite remote sensing and street view imagery data, it further elucidates and enhances the public’s perception of parks. Secondly, it fills a gap in the research related to popularity and emotions by using MLR and Multitask regression to demonstrate the relationships between parks’ ETPN, INPY, GVI, and NDVI. The inclusion of the temporal element further demonstrates how seasons affect the public emotions and INPY of urban parks. Finally, this study provides a new perspective for urban planning and management, helping urban planners identify the factors that influence public user experience, thereby optimizing spatial layouts and facility configurations. Additionally, it proposes management measures for urban parks in different seasons, enhancing the multifunctionality and social benefits of parks to meet the growing public demand, and further improving the livability and sustainability of cities.