How does the level of positive sentiment from park visitors correlate with the number of visitors in POIs nearby parks?
2025 Fall
Intro to Urban Analytics Project
Zhaoxin (Jenny) Ma, Kasturi Palit, Sung Ho Synn
Introduction
Urban parks play a quiet but powerful role in shaping how people move through a city. Beyond their environmental and social benefits, parks create emotional experiences that can influence where people choose to spend time and money. As municipalities increasingly recognize parks as strategic assets for economic stimulation, understanding the mechanisms through which park quality influences local economic vitality becomes essential to effective urban planning policy.
Municipal investments in park infrastructure have demonstrated measurable returns on investment. Across the United States, municipalities are increasingly capitalizing on park infrastructure to stimulate local economic vitality. (National Recreation and Park Association, 2022) For instance, Chattanooga, Tennessee exemplifies this approach through substantial revitalization efforts, including a $15 million riverfront park initiative that generated significant property tax revenue increases. (Local 3 News Staff, 2023) Similarly, Boston has leveraged Privately Owned Public Spaces to combine private capital with public benefit. (City of Boston, 2017) Empirical research confirms these economic impacts: properties proximate to parks command price premiums ranging from 8-10% relative to comparable properties elsewhere in neighborhoods, with larger regional parks showing premiums as high as 16%. (Crompton, 2001) These capitalization effects generate sustained municipal revenue streams through increased property taxation and retail activity.
However, existing research examining parks’ economic impact has focused predominantly on property values and aggregate economic measures. A critical gap remains in understanding how the public’s sentiment and perceived satisfaction with park environments translates into measurable changes in commercial foot traffic and economic activity in adjacent areas. While prior work has established that neighborhood factors influence property values, the relationship between documented visitor experiences and actual patronage patterns at nearby businesses remains underexplored.
Recent advances in data science offer new methodologies to address this research gap. Social media platforms have emerged as rich repositories of expressed sentiment and opinion regarding urban spaces. Recent reserach shows that people’s review ratings about point of interests on social media (Baidu) can enhance understanding and prediction of urban activity intensity. (Wang et al., 2025) Sentiment analysis using deep learning techniques has demonstrated capacity to extract nuanced emotional and evaluative content from user-generated text. Another research utilized twitter texts and employed BERT-based sentiment evaluation model to measuring citizen’s satisfaction on parks across the new york city.(Plunz et al., 2019) Early evidence suggests that integrating sentiment data with structured urban metrics produces more robust predictive models than traditional approaches alone.
Despite these advancements, previous studies have largely relied on limited metrics, such as numerical ratings, or have focused solely on satisfaction with the park itself rather than its spillover effects. There is a distinct lack of research examining whether the sentiments expressed in review texts correlate with the vibrancy of the local economy.
This project explores that connection using Google review sentiment, neighborhood characteristics, and Advan Research foot-traffic data for points of interest (POIs) near 42 parks in Atlanta. By combining sentiment analysis, seasonal POI visit data, and a negative binomial regression framework, the goal is to understand whether positive park experiences translate into measurable increases in nearby economic activity.
Research Questions
RQ1. To what extent does the sentiment of park reviews correlate with the level of foot traffic to points of interest located near those parks?
RQ2. Does the relationship between park sentiment and nearby commercial activity remain significant after accounting for neighborhood and industry characteristics?
RQ3. How does the relationship between park sentiment and nearby commercial activity vary across different seasons (e.g., spring, summer, fall, winter)?
Data
The analysis draws from three primary datasets: park locations and characteristics, Google review text for each park, and seasonal POI visit data.
Atlanta Parks Boundary
Park boundaries were first filtered to include only municipally controlled parks within the City of Atlanta. Extremely large parks such as the BeltLine and Chattahoochee River National Recreation Area were removed because their scale prevents meaningful linkage between localized sentiment and specific nearby businesses. Parks with fewer than five text reviews were also excluded to ensure a reliable sentiment signal.
Google Review
Google review data were collected using web crawling techniques. Each park was geocoded by google maps API to extract its placekey, which was then used to pull the full review page from Google Maps. Using BERT-based sentiment classification, each sentence was scored and filtered by confidence level. This produced a sentiment index for each park that reflects the general emotional tone of user experiences.
The number of visitors to Points of Interest.
Monthly foot traffic data from Advan Research for March (Spring), June (Summer), September (Autumn), and December (Winter) of 2024 was used to capture seasonal variations in visit counts for capturing year-round variability and control for seasonal bias in our model.
The analysis was strictly limited to POIs within the 100-meter buffer of each park. This focus allows for an examination of the park’s influence on adjacent commercial facilities.
Also, the analysis concentrated on industries most relevant to social and recreational activity: Retail, Arts and Entertainment, Accommodation and Food, Wholesale (included mainly as a comparison group).
Sentiment Analysis
Sentiment Analysis Methodology
To evaluate the qualitative aspect of the reviews, we conducted a sentiment analysis using the nlptown/bert-base-multilingual-uncased-sentiment model. This BERT-based classifier is pre-trained on extensive datasets (including Yelp and Amazon reviews), allowing it to effectively capture semantic nuances and context beyond simple word counts.
Scoring and Classification
The model classifies each sentence into a five-point sentiment scale:
- 1 (Very Negative): e.g., “Terrible service”
- 3 (Neutral): e.g., “Okay, not bad”
- 5 (Very Positive): e.g., “Absolutely amazing!”
To ensure the reliability of our analysis, predictions with a model confidence score below 50% were discarded.
Final Sample Size: The analysis was performed on a total of 11,410 sentences.
The distribution of sentiment classes in the final sample is visualized below.
Park Sentiment Score
The final park sentiment score was computed by summing over sentiment scores of sentences divided by the total number of sentences in reviews for each park.
\[ \text{Final Park Score} = \frac{\text{Sum of Sentiment Scores per Sentence}}{\text{Total Number of Sentences in Reviews for Each Park}} \]
Regression
Model Adoption
\[ \ln(\text{Visitor Counts}_i) = \beta_0 + \beta_1(\text{Sentiment Score}_i) + \sum_{k} \gamma_k(\text{Control}_{ki}) \] To estimate the relationship between park sentiment and nearby commercial activity, a negative binomial regression model was used. This approach is well-suited for over-dispersed count data, which characterizes POI visit totals. The model includes sentiment score as the main predictor and controls for a set of socioeconomic and environmental variables known to shape foot traffic patterns. Seasonal models were also estimated separately to test for variation across the year.
Data for Regression Description
To ensure the robustness of the regression model and mitigate potential confounding effects, a comprehensive set of Control Variables was incorporated, categorized into socio-economic characteristics, physical environment, and park attributes:
Socio-economic Factors
To control for the inherent purchasing power and demographic composition that may influence foot traffic independent of park proximity, we included Median Household Income (measured in 2024 USD at the census tract level), Poverty Rate (percentage of population below the poverty level of 1.0), and White Population Rate (percentage of the population identifying as White).
Built Environment & Accessibility
We incorporated several variables to capture the overall urban intensity and infrastructural support. Population and Job Density (per square kilometer) and Intersection Density (count of intersections within a block group) account for urban cluster effects. Street Length (total kilometers of road within a block group) controls for the potential for through-traffic and neighborhood navigability. Crime Density (annual reported crimes per square kilometer) accounts for neighborhood safety perceptions. Furthermore, Transit Availability (measured by the count of bus stops within a block group and the distance to the nearest subway station) and Walk Score control for the non-automobile accessibility of the neighborhood.
Park Characteristics
Finally, Park Area (measured in acres, based on the official park boundary) was included to control for the physical scale and recreational capacity of the park, which is expected to have a direct, non-proximal effect on surrounding visitation.
Descriptive Statistics
The descriptive statistics summarize the key variables used in the regression. Total annual POI visit counts show high variability, reflecting differences in commercial density across park areas. The sentiment score ranges from highly negative to highly positive, with an overall tendency toward positive experiences. Walkability indicators, crime density, intersection density, and built-environment measures show substantial differences across neighborhoods. Socioeconomic variables such as household income, job density, population density, and poverty rate also vary widely, highlighting the diverse context in which Atlanta’s parks operate. These descriptive patterns reinforce the need for a model that accounts for both sentiment and neighborhood characteristics when estimating their influence on POI visits.
Regression Results
| Characteristic |
Model Interpretation
|
Effect Size
|
|||
|---|---|---|---|---|---|
| Coef1 | SE | 95% CI | p-value | IRR1 | |
| sentiment_score | 1.88e+00*** | 0.232 | 1.29e+00, 2.41e+00 | <0.001 | 6.57*** |
| pop_density | -3.24e+01 | 23.0 | -7.22e+01, 1.79e+01 | 0.160 | 0.00 |
| job_density | -1.92e+01* | 8.65 | -3.61e+01, -1.56e+00 | 0.027 | 0.00* |
| hhincome | 4.92e-06* | 0.000 | -8.47e-07, 1.13e-05 | 0.043 | 1.00* |
| poverty_rate | -8.73e-01 | 0.828 | -2.53e+00, 9.79e-01 | 0.292 | 0.42 |
| white_rate | -1.32e+00* | 0.533 | -2.43e+00, -2.39e-01 | 0.013 | 0.27* |
| crime_density | 2.20e-03*** | 0.000 | 1.22e-03, 3.17e-03 | <0.001 | 1.00*** |
| intersection_density_km | -1.73e-02*** | 0.005 | -2.87e-02, -5.65e-03 | <0.001 | 0.98*** |
| total_length_m | -3.78e-05 | 0.000 | -9.34e-05, 2.04e-05 | 0.150 | 1.00 |
| bus_stop_count | -1.26e-02 | 0.014 | -4.25e-02, 1.82e-02 | 0.355 | 0.99 |
| walkscore | -1.34e-02 | 0.008 | -3.05e-02, 2.14e-03 | 0.091 | 0.99 |
| dist_to_subway_m | -2.10e-04* | 0.000 | -4.31e-04, 1.77e-05 | 0.034 | 1.00* |
| Area_Acres | -8.84e-03*** | 0.001 | -1.14e-02, -6.37e-03 | <0.001 | 0.99*** |
| Observations | 302 | ||||
| Log-Likelihood | -2,576 | ||||
| 1 p<0.05; p<0.01; p<0.001 | |||||
| Abbreviations: CI = Confidence Interval, IRR = Incidence Rate Ratio, SE = Standard Error | |||||
The regression results show a consistent and strong relationship between park sentiment and POI visit counts. Higher sentiment scores are associated with significantly greater foot traffic to nearby POIs, even after controlling for neighborhood socioeconomic conditions, physical environment characteristics, and park size. This suggests that what people feel about a park is not only an emotional measure but also a meaningful economic indicator.
Crime density is positively associated with foot traffic, which likely reflects the fact that more active commercial areas naturally generate higher incident reports. Intersection density shows a negative relationship with POI visits, indicating that more complex road networks may create barriers to pedestrian movement or reflect car-oriented environments that dilute local foot activity. Park area shows a negative impact as well, meaning smaller parks may anchor local commercial zones more effectively than large, dispersed green spaces.
Model comparison confirms that adding sentiment significantly improves model fit. The likelihood ratio test shows a meaningful increase in explanatory power when sentiment is included, demonstrating that emotional experience adds information beyond physical or socioeconomic factors alone.
Validation of Model Improvement (ANOVA)
To quantify the explanatory power of the sentiment variable, we performed an ANOVA Likelihood Ratio Test. This method assesses the reduction in residuals achieved by incorporating the sentiment score into the regression equation.
We conducted a comparative analysis between two models: 1. Base Model: Included only control variables (Log-Likelihood: \(-5,182.09\)). 2. Sentiment Model: Included control variables and the Sentiment Score (Log-Likelihood: \(-5,152.44\)).
| Likelihood Ratio Test (Model Comparison) | ||||
| Testing the incremental value of adding Sentiment Score | ||||
| Model | Resid DF | -2 Log-Likelihood | Chi-Square (LR Stat) | P-value |
|---|---|---|---|---|
| Base Model (Controls only) | 289 | −5,182.09 | - | - |
| Sentiment Model (Final) | 288 | −5,152.44 | 29.65 | 5.183 × 10−8 |
The comparison revealed a positive shift in the Log-Likelihood by 29.65. Consequently, the test yielded a p-value of \(< 0.001\), indicating a statistically significant improvement in model fit.
These results confirm that the Sentiment Score provides incremental explanatory power and acts as a crucial predictor rather than a noise variable.
Seasonal Analysis
Seasonal models reveal that the role of sentiment remains surprisingly stable throughout the year. Although winter exhibits the best model performance and lowest AIC, possibly because cold-weather behavior becomes more intentional and destination-driven, the effect of sentiment remains consistently positive and statistically significant in all four seasons. The overlapping confidence intervals demonstrate that sentiment is a reliable year-round predictor of commercial activity near parks.
Industry-Specific Analysis
| Characteristic |
Arts & Entertainment (71)
|
Food & Accommodation (72)
|
Retail (44-45)
|
|||
|---|---|---|---|---|---|---|
| log(IRR)1 | SE | log(IRR)1 | SE | log(IRR)1 | SE | |
| sentiment_score | 1.9*** | 0.518 | 1.6*** | 0.297 | 2.2*** | 0.416 |
| dist_to_subway_m | 0.00 | 0.000 | 0.00*** | 0.000 | 0.00 | 0.000 |
| crime_density | 0.00** | 0.001 | 0.00* | 0.000 | 0.00** | 0.000 |
| intersection_density_km | -0.02 | 0.011 | -0.01 | 0.006 | -0.03*** | 0.009 |
| hhincome | 0.00 | 0.000 | 0.00* | 0.000 | 0.00** | 0.000 |
| white_rate | -0.15 | 1.26 | -1.3 | 0.717 | -2.2** | 0.838 |
| Area_Acres | -0.01** | 0.003 | -0.01** | 0.002 | -0.01*** | 0.002 |
| 1 p<0.05; p<0.01; p<0.001 | ||||||
| Abbreviations: CI = Confidence Interval, IRR = Incidence Rate Ratio, SE = Standard Error | ||||||
Breaking down the results by industry shows the universal importance of sentiment across arts, food, and retail categories. Retail shows the largest coefficient, suggesting that consumer-oriented businesses are particularly sensitive to the atmosphere created by nearby parks. The stability of sentiment across industries reinforces its value as a general indicator of localized economic vitality.
Additional Analysis
Frequently Mentioned Words
Beyond the regression, text mining of review content reveals the themes that drive positive and negative sentiment. Negative reviews frequently highlight cleanliness and safety concerns, while positive reviews emphasize amenities, atmosphere, and recreational experiences. These patterns align with the statistical results: management and experiential qualities appear to influence how people feel about parks and, in turn, how they navigate the surrounding commercial landscape.
Analysis by Park Categories
Following the establishment of a statistically significant correlation between park area and the number of Points of Interest (POI) visits, a comprehensive analysis was conducted on the distribution of park types and the visitation tendency of POI visitors across park categories defined by their size (area) within Atlanta.
Analyses of park categories also show that tiny and medium parks consistently support strong commercial activity, while large parks are less associated with concentrated POI visits. This supports the idea that compact parks embedded in dense neighborhoods may deliver higher localized economic benefits.
Discussion and Implications
The findings highlight the importance of emotional experience in shaping urban economic patterns. Park sentiment has a measurable and stable link to nearby POI activity, emphasizing that people respond not only to physical infrastructure but also to the atmosphere and “vibe” of public spaces. Small, high-quality parks appear to act as reliable economic anchors, supporting retail and food businesses throughout the year.
For planners and city officials, these results suggest that investing in park maintenance, programming, and user experience may generate stronger commercial outcomes than simply expanding park acreage. Improving the quality of small parks, enhancing cleanliness and safety, and adding amenities could help strengthen local economies and support walkable urban centers.
Limitations
This study relies on Google review sentiment, which may not fully represent all park users. POI visits are measured within a fixed 100-meter buffer, which may not capture broader walking patterns or the influence of larger commercial districts. Seasonal sampling is limited to four months, and mobility trace data beyond SafeGraph were not incorporated. Additionally, sentiment may reflect broader neighborhood conditions, creating potential endogeneity in interpretation.
Future Research
Future work could track sentiment longitudinally to capture changes related to events, renovations, or seasonal programming. Expanding the analysis to compare cities would test whether the Atlanta patterns generalize to other urban contexts. Integrating mobility data, walksheds, and transit accessibility could provide a richer understanding of pedestrian flow. Sensitivity testing with different buffer distances would also strengthen the robustness of the results.
Conclusion
This project shows that how people feel about a park matters for local economy in a very real and measurable way. Positive park sentiment is strongly connected to higher foot traffic at nearby points of interest, even after accounting for neighborhood conditions, built-environment factors, and park size. The results suggest that the emotional experience of being in a park can influence the choices people make about where to go next, especially for retail and food-related businesses.
One of the clearest takeaways is that small, well-maintained parks can have an outsized influence on the local economy. These compact parks seem to act as steady anchors for surrounding businesses across all seasons, while larger parks show less concentrated commercial impact. The seasonal models also show that the role of sentiment stays stable throughout the year, which means that sentiment can be treated as a reliable, year-round indicator rather than something that fluctuates significantly with weather or timing.
Another important insight comes from the review text itself. People respond strongly to issues like cleanliness, safety, and available amenities. These management-related factors line up closely with what the regression results highlight, reinforcing the idea that improving the quality and experience of a park can benefit the businesses around it.
Overall, the findings point toward a simple but powerful idea: improving park experience may be just as valuable as physical expansion when it comes to supporting local economic vitality. Investments that boost satisfaction—better maintenance, more amenities, or thoughtful programming—could strengthen the link between parks and surrounding commercial activity. As cities continue thinking about how to create lively, people-centered districts, this relationship between sentiment and POI visits is an important piece of the puzzle.
Reference
- National Recreation and Park Association. (2022). The economic impact of local parks. https://www.nrpa.org/publications-research/research-papers/the-economic-impact-of-local-parks/
- Local 3 News Staff. (2023, April 19). Tennessee investing $15 million to help Chattanooga Riverfront parks flourish. Local 3 News. https://www.local3news.com/local-news/tennessee-investing-15-million-to-help-chattanooga-riverfront-parks-flourish/article_4bdd2c02-058c-4a8c-bf00-0813c17832cd.html
- City of Boston. (2017). Imagine Boston 2030: A plan for the future of Boston. https://imagine.boston.gov
- Wang, L., Wang, G., Liu, Z., & Ma, R. (2025). Exploring urban activity intensity with Gravity-Based Betweenness Centrality and social media evaluation: A case study of Beijing. Sustainable Cities and Society, 134, 106898. https://doi.org/10.1016/j.scs.2025.106898
- Plunz, R. A., Zhou, Y., Carrasco Vintimilla, M. I., Mckeown, K., Yu, T., Uguccioni, L., & Sutto, M. P. (2019). Twitter sentiment in New York City parks as measure of well-being. Landscape and Urban Planning, 189, 235–246. https://doi.org/10.1016/j.landurbplan.2019.04.024