Analyzing the relationship between urban network centrality and third place characteristics of bookstores in Brooklyn and Queens, NYC.

This project uses computational methods to measure whether a bookstore's position in the street network predicts its function as a "third place": a social space outside home and work where people gather informally.

Key Finding: Network centrality does not predict third place characteristics (r = -0.15, p = 0.24), but distinct bookstore archetypes emerge with clear geographic clustering patterns.

1. Network Analysis: Extract street networks for Brooklyn & Queens from OpenStreetMap, compute closeness and betweenness centrality for all nodes
2. Bookstore Data: Collect bookstore locations from OSM, map to nearest network nodes
3. Review Collection: Gather Google Places reviews for all bookstores, filter out closed locations
4. LLM Classification: Use Qwen 2.5 14B to classify each review against 8 third place dimensions from Langlais & Vaux (2022)
5. Aggregation: Calculate per-location scores (proportion of reviews mentioning each dimension)
6. Analysis: Correlate centrality metrics with third place scores, derive composite archetypes

Based on Langlais & Vaux (2022) framework:

• Relationship Initiation & Maintenance
• Equalizer (welcoming to all)
• Communication as Main Activity
• Active & Passive Engagement
• Reciprocity (regulars, recognition)
• People Over Place
• Playful Mood
• Cognitive Separation & Reprieve

Three distinct types emerged from dimension patterns:

Sanctuary (quiet refuge for cognitive separation)

• High: cognitive_separation, equalizer
• Low: communication_activity
• Examples: Head Hi, Kew & Willow Books, Greenlight Bookstore

Social Hub (buzzing conversation and social energy)

• High: communication_activity, relationship_initiation, playful_mood
• Examples: Books Are Magic, Center for Fiction Bookstore

Local Living Room (intimate space for neighborhood regulars)

• High: reciprocity, people_over_place, equalizer
• Rare but distinctive pattern
• Examples: Taylor & Co., Eichler's Religious Articles

1. No correlation between network centrality and third place function

   • Closeness centrality: r = -0.151, p = 0.241
   • Betweenness centrality: r = -0.122, p = 0.343
   • No individual dimensions showed significance

2. Strong geographic clustering

   • Northwest Brooklyn (Williamsburg, Greenpoint, Park Slope areas) shows highest concentration of Sanctuaries and Social Hubs
   • Local Living Room archetype is rare across all neighborhoods
   • Suggests neighborhood context matters more than street network position

3. Sample characteristics

   • 60 bookstores analyzed
   • 291 total reviews
   • Average 4.8 reviews per location
   • Aggregate third place scores: mean 0.30, range 0.00-0.63

• Borough boundaries: NYC Open Data (included in data/shapefiles/)
• Street networks: Generated via OSMnx
• Bookstore locations: OpenStreetMap features
• Reviews: Google Places API
• Final outputs: data/processed/ (centrality scores, location scores)

1. Clone repository
2. Create .env file with Google Places API key:

GOOGLE_PLACES_API_KEY=your_key_here

3. Install Ollama and pull Qwen model:

ollama pull qwen2.5:14b

4. Setup Python virtual environment and install requirements:

pip install -r requirements.txt

Run in order:

1. 01-data-prep.ipynb - Download networks and bookstore locations
2. 02-centrality-metrics.ipynb - Calculate network centrality
3. 03-collect-reviews.ipynb - Fetch Google Places reviews
4. 04-classify-bookstores.ipynb - LLM classification of bookstores based on reviews
5. 05-analysis.ipynb - Correlation analysis and visualization

Interactive maps available in figures/:

• bookstore_map.html - Overall third place scores
• archetype_sanctuary.html - Sanctuary archetype distribution
• archetype_social_hub.html - Social Hub archetype distribution
• archetype_local_living_room.html - Local Living Room archetype distribution

• Review-based measurement may underrepresent certain characteristics (e.g., reciprocity rarely mentioned explicitly)
• Sample size (60 bookstores) limited to active locations with sufficient reviews
• LLM classification dependent on review text quality and prompt engineering
• Cross-sectional analysis; no temporal component

Rahman, M. (2025). Third Place Persona: Measuring Social Characteristics of Bookstores. 
GitHub: github.com/msradam/third-place-persona

Langlais, M. R., & Vaux, D. E. (2022). Establishing and Testing a Quantitative Measure for Evolving Third-Place Characteristics. International Journal of Technology and Human Interaction, 18(1), 1-15.

MIT License Copyright (c) 2025 Adam Munawar Rahman