Evaluating the Influence of Park Attributes on User Satisfaction Using Flickr Data: A Case Study of Southeast Michigan

Jie YangJournal of Humanities Arts and Social Science 2025 · DOI

TL;DR

This paper leverages Flickr data and sentiment analysis to understand how specific park features influence user satisfaction in Southeast Michigan. The findings highlight the importance of natural elements and historical sites for positive experiences, offering actionable insights for urban planners to design more engaging and satisfying green spaces.

What Problem Does This Paper Address?

Traditional methods for assessing public satisfaction with urban parks, such as surveys, are often costly and time-consuming. This research addresses the need for more efficient and real-time methods by exploring the potential of social media data. Specifically, it aims to understand how various park attributes (e.g., natural areas, recreational facilities) impact user sentiment and satisfaction, thereby informing better urban park design and planning. The study seeks to bridge the gap between digital social data and tangible urban planning decisions.

Methods

The study employs a multi-faceted approach:

  • Data Collection: Flickr data (geo-tagged photos and user comments) was collected for parks in Southeast Michigan. This was cross-referenced with park attribute data from the Southeast Michigan Council of Governments (SEMCOG).
  • Data Preprocessing: Complementary data from Google Earth and OpenStreetMap were used to verify geographic accuracy. Natural Language Processing (NLP) techniques were applied to clean and preprocess textual descriptions, removing duplicates and irrelevant content.
  • Sentiment Classification: User sentiment was categorized into "Irrelevant," "Pro-nature (positive sentiment)," "Pro-nature (negative sentiment)," and "Against-nature (negative sentiment towards park environment)." A sentiment score from 0 to 1 was assigned to each comment. The study utilized advanced NLP and machine learning techniques, including BERT (Bidirectional Encoder Representations from Transformers) and ensemble learning, to enhance classification accuracy by considering word context and combining multiple classifiers.
  • Regression Analysis: A linear regression model was developed to quantify the relationship between 46 park attributes (independent variables) and the normalized sentiment score (dependent variable). Park size was included as a dummy variable. LASSO regression was used to reduce multicollinearity, and a 10-fold cross-validation technique was employed for model validation.

Datasets

  • Flickr Data: Geo-tagged photos and user comments from Flickr, collected over a 15-year period (2007-2022). Approximately 3,063 valid comments were analyzed.
  • Park Attribute Data: Information on park attributes for 2,657 parks in Southeast Michigan, obtained from the Southeast Michigan Council of Governments (SEMCOG).
  • Geographic Data: Google Earth and OpenStreetMap data were used to verify geographic accuracy of Flickr data.

Key Insights and Ideas

  • The Power of Nature and History: The study strongly suggests that natural areas and historical features are key drivers of positive user experiences in parks. This aligns with the intuitive understanding that people seek tranquility and connection to heritage in green spaces.
  • Recreational Facilities: A Mixed Bag: While some recreational facilities are appreciated, others, like BMX areas and equestrian activities, were found to negatively impact user satisfaction. This implies a need for careful consideration of the target audience and potential conflicts when designing such facilities.
  • The Importance of Accessibility and Maintenance: Beyond inherent features, the study points to the significance of accessibility (proximity to urban centers, ease of public transport) and maintenance in shaping user sentiment. Well-maintained parks with accessible features contribute to higher satisfaction.
  • Social Data as a Planning Tool: The research demonstrates that social media analytics, particularly from platforms like Flickr, can provide valuable, cost-effective, and real-time insights for urban planning and landscape design, complementing traditional methods.

Key Results

  • Sentiment Distribution: Out of 3,063 analyzed Flickr comments, 1,808 were classified as pro-nature (positive sentiment), 53 as pro-nature (negative sentiment), and 1,202 as irrelevant. No comments were classified as "against-nature."
  • Significant Park Attributes (at 0.05 significance level):
    • Positive Impact: Natural areas (p=0.000), Historical features (p=0.000), Paved walking/biking trails (p=0.003), Picnic shelters (p=0.011).
  • Negative Impact: BMX areas (p=0.041), Equestrian activities (p=0.048).
  • Spatial Analysis: Parks in urbanized areas with greater accessibility showed higher positive sentiment scores.
  • Qualitative Findings: Frequently mentioned positive terms in highly rated parks included "beautiful," "peaceful," and "well-maintained."

Limitations and Future Work

Limitations:

  • Limited Flickr Comments: The number of Flickr comments may affect the accuracy of the results.
  • Data Imbalance: Some parks received significantly more user input than others, creating data imbalance.
  • Context and Sarcasm: Sentiment analysis can struggle with accurately interpreting context, sarcasm, and linguistic nuances, necessitating continuous improvement of NLP models.
  • Demographic Bias: Social data may not accurately reflect the entire population, potentially skewing demographic representation.
  • Privacy Concerns: Ethical considerations regarding privacy and user consent for using social data are important.

Future Work:

  • Multimodal Analysis: Integrate data from other platforms like Instagram, Twitter, and Google Reviews, and combine text, image, and user check-in data for a more comprehensive assessment.
  • Advanced ML Techniques: Apply more sophisticated analytical techniques, including deep learning and reinforcement learning, for enhanced sentiment classification and dynamic park design adjustments.
  • Geospatial Modeling: Further explore geospatial regression models to examine the spatial distribution of sentiment scores and their correlation with land use patterns.
  • Environmental Factors: Investigate the role of environmental factors like air quality and noise levels.
  • Qualitative Integration: Combine social analytics with qualitative methods (focus groups, interviews) for richer insights.
  • Predictive Modeling: Utilize big data analytics for predictive modeling of visitor patterns and resource allocation.
  • Emerging Technologies: Explore augmented reality (AR) and virtual reality (VR) for testing design scenarios and smart infrastructure for enhanced user engagement.

Summary updated

Abstract

The integration of social data in urban studies presents new opportunities for assessing public perceptions of green spaces. This research investigates how park attributes influence visitor sentiment using Flickr data from Southeast Michigan. By analyzing geo-tagged photos and user comments, this study applies sentiment analysis and linear regression modeling to identify the key features that enhance or detract from user satisfaction. The results indicate that natural areas, historical monuments, and walking trails significantly improve user experiences, whereas BMX areas and equestrian activities tend to reduce satisfaction levels. These findings suggest that urban planners and landscape architects should emphasize natural elements and cultural landmarks in park design to foster positive user engagement. Future research should integrate data from multiple social platforms and apply more advanced machine-learning techniques to further validate these insights. This study underscores the potential of social analytics as a valuable tool for urban planning and landscape development.
BibTeX
@article{yang2025evaluating,
  author = {Jie Yang},
  title = {{Evaluating the Influence of Park Attributes on User Satisfaction Using Flickr Data: A Case Study of Southeast Michigan}},
  journal = {Journal of Humanities Arts and Social Science},
  year = {2025},
  doi = {10.26855/jhass.2025.03.001},
}
Figure