Analyzing Urban Vitality at scale: PhD defense 2019

Transcript

1. Marco De Nadai, supervised by Bruno Lepri and Nicu Sebe
   INTO THE CITY:
   a Multi-Disciplinary
   Investigation of Urban Life
   

   View Slide

2. 2
   Cities, very difficult to explain
   Not only agglomeration of
   residents, factories, shops
   • Millions of individuals
   • Continuously evolving
   A small change generates a
   cascading throughout
   COMPLEX SYSTEM
   

   View Slide

3. 3
   Cities have always been studied
   IDEAL CITY
   (XV century)
   SYSTEM
   (XIX century)
   LIVING ORGANISM
   (XX century)
   

   View Slide

4. 4
   Understand cities
   

   View Slide

5. 5
   Understand cities
   New methods
   New data
   

   View Slide

6. 6
   Describe people and places at scale
   Urban
   description
   1
   Urban
   perception
   2
   

   View Slide

7. 7
   CITY PEOPLE
   Multi-modal approach
   

   View Slide

8. Urban description
   Q: Can we describe how people experience the city?
   1
   

   View Slide

9. The theory: Jane Jacobs
   One of the most influential books in city
   planning
   • planning models that dominated mid-
   century planning
   • Melbourne, Toronto etc.
   9
   Klemek, C. (2011) ‘Dead or Alive at Fifty? Reading Jane Jacobs
   on her Golden Anniversary’ Dissent, Vol. 58, No. 2, 75–79.
   2
   1 3
   URBAN DESCRIPTION 4
   

   View Slide

10. The theory: Jane Jacobs
    • Written in 1961
    • Not empirically tested until 2015
    • Tested in Seoul, from costly surveys
    collected in years
    • Operationalize the theory
    10
    Jacobs, Jane. The death and life of great American cities.
    Vintage, 1961
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

11. The theory: Jane Jacobs
    The theory says that:
    • Death: caused by the elimination of
    pedestrian activity
    • Life: created by a vital urban fabric at
    all times of the day
    11
    Jacobs, Jane. The death and life of great American cities.
    Vintage, 1961
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

12. The theory: Jane Jacobs
    Diversity => Urban vitality
    There are 4 diversity conditions
    12
    LAND USE
    SMALL
    BLOCKS
    AGED
    BUILDINGS
    DENSITY
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

13. Operationalize the theory
    • Land use mix
    13
    For district :
    %
    = − (
    )∈+
    %,)
    log(%,)
    )
    log ||
    %,)
    : % square footage of land use
    : {residential, commercial, recreation}
    LAND USE
    SMALL
    BLOCKS
    AGED
    BUILDINGS
    DENSITY
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

14. Operationalize the theory
    • Small blocks
    |%
    |
    %
    14
    LAND USE
    SMALL
    BLOCKS
    AGED
    BUILDINGS
    DENSITY
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

15. Operationalize the theory
    • Aged buildings:
    15
    @AB
    = ∑DEF
    G HD(IDJ ̅
    I)L
    M
    (NOF)
    N
    ∑DEF
    G HD
    =
    LAND USE
    SMALL
    BLOCKS
    AGED
    BUILDINGS
    DENSITY
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

16. Operationalize the theory
    16
    LAND USE
    SMALL
    BLOCKS
    AGED
    BUILDINGS
    DENSITY
    2
    1 3 4
    Employment density:
    |PQRSTUBV RBTRSBD|
    @WB@D
    Population density:
    |XTRYS@Z%T[D|
    @WB@D
    URBAN DESCRIPTION
    

    View Slide

17. Operationalize the theory
    • 6 Italian cities
    • Features to describe the Jane Jacobs
    theory
    17
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

18. What is vitality?
    • Defined in various (fuzzy) ways in
    urban science and sociology
    • There is no standard
    • Key asset for urban spaces
    • Important for companies (and retail)
    success
    • Influences the real estate market
    18
    asd
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

19. Call Detail Records
    Data collected by mobile operators for
    billing reasons
    • Unique userID
    • Gender and age
    • Geographical location (Antenna)
    • Datetime
    19
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

20. Vitality (empirical)
    • Define areas
    • We define vitality as the average
    number of people in a neighborhood
    20
    1
    %
    ||
    (
    _∈`
    | ℎ %
    |
    : set of hours (60 days x 24h)
    : area of district
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

21. Describe urban areas and vitality
    21
    VITALITY
    For each neighborhood
    Mobile data
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

22. Describe urban areas and vitality
    22
    People + Companies
    GIS
    VITALITY
    For each neighborhood
    Mobile data
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

23. Describe urban areas and vitality
    23
    People + Companies
    GIS
    Predictive
    model
    VITALITY
    For each neighborhood
    Mobile data
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

24. The log Linear Regression model
    24
    Vitality (Ground truth)
    Land Use Mix
    Employment density
    = i
    i
    + l
    l
    + ⋯ + [
    [
    +
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

25. 25
    Urban metric Standardized Beta coefficient
    Employment density 0.434***
    Intersections density 0.191***
    Housing types 0.185***
    Closeness highways -0.102***
    3rd places x closeness highways 0.07**
    Closeness parks x closeness highways -0.07***
    adj − Rl 0.77
    *** p-value < 0.001; ** p-value < 0.01;
    Describe urban vitality
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

26. Take home
    26
    De Nadai, Marco, et al. "The Death and Life of Great Italian Cities: A Mobile Phone Data Perspective."
    WWW, 2016.
    Physical characteristics
    describe and predict
    urban vitality
    2
    1 3 4
    URBAN DESCRIPTION
    

    View Slide

27. Urban perception
    Q: Can we link urban visual perception with the behavior of
    people at scale?
    2
    

    View Slide

28. Broken windows theory
    • City mismanagement
    • Dirty places
    • Poor infrastructure
    Lead to misbehavior => Crime
    Q: Are people avoiding places where
    they feel unsafe?
    28
    Wilson, James Q., and George L. Kelling. "Broken windows."
    Critical issues in policing: Contemporary readings (1982): 395-
    407.
    2
    1 3
    URBAN PERCEPTION 4
    

    View Slide

29. 29
    Urban perception from Place Pulse
    Salesses, P., Schechtner, K., & Hidalgo, C. A. (2013). The collaborative image of the city: mapping the inequality of urban
    perception. PloS one
    2
    1 3
    URBAN PERCEPTION 4
    

    View Slide

30. 30
    …
    1
    10
    Place Pulse
    • New York
    • Boston
    • Linz
    • Salzburg
    Place Pulse 2
    • Rome
    • Milan
    PROBLEM:
    • Few images per neighborhood
    • Few labels per image
    2
    1 3
    URBAN PERCEPTION 4
    Safety perception: MIT Place Pulse
    

    View Slide

31. 31
    Security perception prediction
    * B. Zhou, A. Lapedriza, J. Xiao, A. Torralba, and A. Oliva. “Learning Deep Features for Scene
    Recognition using Places Database.” NIPS, 2014.
    • Learning human security perception
    • Transfer learning from Place205*, US, to Rome and Milan
    PERCEPTION
    SCORE
    [0-10]
    2
    1 3
    URBAN PERCEPTION 4
    

    View Slide

32. 32
    Can we predict human security perception?
    * Ordonez, Vicente, and Tamara L. Berg. "Learning high-level judgments of urban perception.” ECCV, 2014.
    Model type State of the art* Our model
    NY - NY 0.687 0.718
    NY - Boston 0.701 0.734
    Boston - Boston 0.718 0.744
    Boston - NY 0.636 0.693
    2
    1 3
    URBAN PERCEPTION 4
    

    View Slide

33. 33
    2
    1 3
    URBAN PERCEPTION 4
    Describe security perception
    Security perception Vitality
    Regression model
    

    View Slide

34. 34
    2
    1 3
    URBAN PERCEPTION 4
    Describe security perception
    2 4.4 4.6 4.8 5.0
    ety score
    DUOMO
    SAN SIRO
    QUARTO
    OGGIARO
    CITTA'
    STUDI
    BICOCCA
    0.2
    0.0
    0.2
    0.4
    0.6
    0.8
    1.0
    1.2
    Land Use Mix
    0.8 1.2 1.8 2.7 4.1 6.1 9.3
    Activity density ⇥ 10
    MILAN
    Security perception Vitality
    Regression model
    

    View Slide

35. 35
    Urban metric Standardized Beta coefficient
    adj − Rl 0.91
    Security perception -> presence of people
    2
    1 3
    URBAN PERCEPTION 4
    

    View Slide

36. 36
    Urban metric Standardized Beta coefficient
    Population density 0.155**
    Employees density 0.328**
    Deprivation -0.022
    Distance from the center -0.257**
    Security perception 0.105**
    adj − Rl 0.91
    ** p-value < 0.001; * p-value < 0.01;
    Security perception -> presence of people
    2
    1 3
    URBAN PERCEPTION 4
    

    View Slide

37. 37
    Urban metric Standardized Beta coefficient
    % of women (from census) 0.001
    Deprivation -0.005
    Distance from the center -0.003
    Security perception 0.020**
    adj − Rl 0.65
    ** p-value < 0.001; * p-value < 0.01;
    Security perception -> presence of women
    2
    1 3
    URBAN PERCEPTION 4
    

    View Slide

38. 38
    Visual elements for security perception
    HIGH SAFETY PERCEPTION
    RANDOMLY OBSCURE
    PART OF THE IMAGE
    AND PREDICT
    2
    1 3
    URBAN PERCEPTION 4
    CONTRIBUTE POSITIVELY
    CONTRIBUTE NEGATIVELY
    

    View Slide

39. Take home
    39
    De Nadai, Marco, et al. "Are Safer Looking Neighborhoods More Lively?: A Multimodal Investigation
    into Urban Life." ACM MM, 2016.
    Security perception can
    describe and predict
    the presence of people
    2
    1 3
    URBAN PERCEPTION 4
    

    View Slide

40. 40
    Describe the places at scale, automatically
    Urban
    description
    1
    Urban
    perception
    2
    

    View Slide

41. Applications
    

    View Slide

42. Housing price
    Q: Can we describe the neighborhood effect on housing price?
    3
    

    View Slide

43. Real estate appraisal
    • Timeless sale transactions
    • Proprietary data
    • Lack of data?
    • Neighborhood?
    43
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

44. A data mining approach
    44
    PLACE
    Characteristics of the
    census cell
    NEIGHBORHOOD
    Description and perception
    PROPERTY
    Characteristics of the
    property
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

45. 45
    The data
    One year of listed properties in Immobiliare.it
    • The 8 biggest Italian cities
    • 70,000 properties
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

46. 46
    The property: textual features
    Textual features about the listed property
    • Number of rooms
    • Square meters
    • Energy compliance
    • Garden (yes/no)
    • […]
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

47. We need context – auto-correlation
    47
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

48. The egohoods
    48
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

49. Features of the city
    For each of the ~ 400’000 census cells:
    • build the egohood
    • Create:
    • 10 socio-economic indexes
    • 11 urban features
    • 4 indexes of companies and jobs
    49
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

50. 50
    XGBoost
    K-fold Cross-validation (with care!)
    = ( + )
    2
    1 3
    HOUSING PRICE 4
    Egohood features
    (e.g. land use mix)
    Property features
    (e.g. square feets)
    Housing price
    (ground truth)
    Weight/contiguity matrix
    

    View Slide

51. 51
    Results
    Model MAE MdAPE
    Property 148, 109 23,76%
    Property + Neighborhood 104,586 15,44%
    THE NEIGHBORHOOD SHAPES
    PROPERTY PRICE BY ~60%!
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

52. 52
    Global features importance
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

53. 53
    Global features importance
    THE NEIGHBORHOOD IS VERY
    IMPORTANT FOR THE PROPERTY PRICE
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

54. Result
    54
    De Nadai, Marco and Bruno Lepri. "The economic value of neighborhoods: Predicting real estate
    prices from the urban environment" IEEE DSAA, 2018.
    Neighborhood features are
    very correlated
    with housing price
    2
    1 3
    HOUSING PRICE 4
    

    View Slide

55. Crime
    Q: Can we describe how the physical characteristics might
    influence crime?
    4
    

    View Slide

56. What drives crime?
    There are many theories
    CRIMINOLOGY
    • Lack of cooperation and trust
    URBAN PLANNING
    • Lack of informal surveillance:
    (guardianship by ordinary citizen, not
    just the police)
    56
    2
    1 3
    CRIME 4
    

    View Slide

57. Limits
    • Mobility of people is not considered
    • The built environment?
    • Usually tested in one city
    Q: Can we study multiple factors in
    multiple cities to understand crime?
    57
    2
    1 3
    CRIME 4
    

    View Slide

58. 58
    2
    1 3
    CRIME 4
    SOCIO-ECONOMIC
    Our model
    • Socio-economic conditions (CRIMINOLOGY)
    • Economic deprivation
    • Ethnic heterogeneity
    • Residential instability
    

    View Slide

59. 59
    2
    1 3
    CRIME 4
    Our model
    • Socio-economic conditions (CRIMINOLOGY)
    • The built environment (URBAN PLANNING)
    • Land use mix
    • Small blocks
    • […]
    BUILT
    ENVIRONMENT
    

    View Slide

60. 60
    2
    1 3
    CRIME 4
    Our model
    • Socio-economic conditions (CRIMINOLOGY)
    • The built environment (URBAN PLANNING)
    • Mobility of people
    MOBILITY
    Jiang, Shan, et al. "The TimeGeo modeling framework for urban mobility without
    travel surveys." Proceedings of the National Academy of Sciences 113.37 (2016)
    

    View Slide

61. 61
    2
    1 3
    CRIME 4
    Our model
    • Socio-economic conditions (CRIMINOLOGY)
    • The built environment (URBAN PLANNING)
    • Mobility of people
    • Tested for Bogotá, Boston, Chicago, Los Angeles
    

    View Slide

62. Describe crime
    62
    Socio-economic
    Built environment
    Descriptive
    model
    CRIME NUMBER
    For each neighborhood
    2
    1 3
    CRIME 4
    Mobility
    

    View Slide

63. 63
    Leroux et al. "Estimation of disease rates in small areas: a new mixed model for spatial dependence."
    Statistical models in epidemiology, the environment, and clinical trials.
    log %
    = (
    vwi
    [
    v
    v
    + CAR process
    Auto-correlation matrix
    Features
    (e.g. land use mix)
    Crime
    in a district
    (ground truth)
    Bayesian Poisson model
    2
    1 3
    CRIME 4
    

    View Slide

64. Results (MdAPE errors)
    64
    Model Bogota Boston Los Angeles
    Socio-economic 44% 43% 22%
    Built environment 24% 31% 22%
    Mobility 37% 40% 21%
    2
    1 3
    CRIME 4
    

    View Slide

65. Results (MdAPE errors)
    65
    Model Bogota Boston Los Angeles
    Socio-economic 44% 43% 22%
    Built environment 24% 31% 22%
    Mobility 37% 40% 21%
    Full model 19% 38% 15%
    SOCIO-ECONOMIC + BUILT ENVIRONMENT + MOBILITY
    2
    1 3
    CRIME 4
    

    View Slide

66. 66
    Small blocks
    Just an example…
    2
    1 3
    CRIME 4
    Built environment - Discrepancies
    

    View Slide

67. BOSTON
    BOGOTA
    67
    Just an example…
    2
    1 3
    CRIME 4
    Built environment - Discrepancies
    

    View Slide

68. BOSTON
    BOGOTA
    68
    THE BUILT ENVIRONMENT IS NOT ENOUGH
    Just an example…
    2
    1 3
    CRIME 4
    Built environment - Discrepancies
    

    View Slide

69. 69
    How can we describe crime?
    Socio-economic
    characteristics
    Built environment Mobility
    ✓ Precise data ✓ Availability
    ✓ Unbiased
    (population)
    ✗ Rarely updated
    ✗ Availability
    ✗ Bias over OSM volunteers ✗ Availability
    2
    1 3
    CRIME 4
    

    View Slide

70. 70
    How can we describe crime?
    Socio-economic
    characteristics
    Built environment Mobility
    ALL COMPONENTS TOGETHER
    BETTER DESCRIBE CRIME
    THEORY HAS DISCREPANCIES
    OVER DIFFERENT CITIES
    2
    1 3
    CRIME 4
    

    View Slide

71. 71
    Research themes
    Urban
    description
    1 4
    2 3
    Urban
    perception
    Housing price Crime
    

    View Slide

72. 72
    Research themes
    Urban
    description
    1 4
    2 3
    Urban
    perception
    Housing price Crime
    AUTOMATICALLY COLLECTED
    DATA
    URBAN SCIENCE AT SCALE
    

    View Slide

73. Why does it matter?
    

    View Slide

74. Why does it matter?
    74
    • DATA MINING
    • Inexpensive way to understand urban mechanisms;
    • New stimulus to social research;
    • Responsive predictions without historical data
    • Deep understanding of city life through multi-modal data
    • Studying cities means studying people
    • Gentrification?
    

    View Slide

75. Some limits
    75
    • Theories/models <-> domain adaptation
    • Timely predictions
    • Data driven results <-> decisions
    

    View Slide

76. 76
    In the (next) future
    DATA DRIVEN APPROACH: let the data speak
    STREET VIEW
    IMAGERY
    AERIAL
    IMAGERY
    + NEIGHBORHOOD
    OUTCOMES
    (CRIME,
    VITALITY…)
    

    View Slide

77. 77
    In the (next) future
    GANs FOR URBAN PLANNING (with Yahui Liu)
    CRIME
    VITALITY
    HOUSING PRICE
    PREDICT
    GENERATE?
    NEIGHBORHOOD
    OUTCOMES
    AERIAL + STREET VIEW
    IMAGERY
    

    View Slide

78. 78
    1. De Nadai, M., Staiano, J., Larcher, R., Sebe, N., Quercia, D., & Lepri, B. The death and life of great Italian
    cities: a mobile phone data perspective. WWW 2016.
    2. De Nadai, M., Vieriu, R. L., Zen, G., Dragicevic, S., Naik, N., Caraviello, M., ... & Lepri, B. (2016, October).
    Are safer looking neighborhoods more lively?: A multimodal investigation into urban life. ACM MM
    2016.
    3. De Nadai, M., & Lepri, B. The economic value of neighborhoods: Predicting real estate prices from the
    urban environment. IEEE DSAA 2018.
    4. De Nadai, M., & Lepri, B. (2018, October). Socio-economic, built environment, and mobility condi-
    tions associated with crime: A study of multiple cities. Under submission to Nature Human
    Behaviour, 2019.
    The topics of this thesis
    

    View Slide

79. 79
    5. Barlacchi, Gianni, et al. "A multi-source dataset of urban life in the city of Milan and the Province of
    Trentino." Nature Scientific Data 2, 2015.
    6. Centellegher, Simone, et al. "The Mobile Territorial Lab: a multilayered and dynamic view on parents’
    daily lives." EPJ Data Science 5.1, 2016.
    7. Mamei, Marco, et al. "Is social capital associated with synchronization in human communication? An
    analysis of Italian call records and measures of civic engagement." EPJ Data Science 7.1, 2018.
    8. De Nadai, Marco, et al. "Apps, Places and People: strategies, limitations and trade-offs in the physical
    and digital worlds." under review in Nature Scientific Reports, 2019.
    9. Strano, Emanuele, et al. "Precise mapping, density and spatial structure of all hu- man settlements on
    earth», under submission for Nature Communications, 2019.
    10. Liu, Yahui, et al. "Gesture-to-gesture translation in the wild via category-independent con- ditional
    maps", under review in ACM MM, 2019.
    Other topics I had the chance to explore
    

    View Slide

80. STEFAN
    DRAGICEVIC
    LETOUZE’
    EMMANUEL
    DANIELE QUERCIA
    MARTA C.
    GONZALEZ
    CESAR A.
    HIDALGO
    JACOPO STAIANO
    NICU SEBE
    BRUNO LEPRI
    ROBERTO
    LARCHER
    SANDY PENTLAND
    GLORIA ZEN
    XU YANYAN
    RADU L. VIERIU
    NIKHIL NAIK THANKS
    

    View Slide

81. Thank you
    

    View Slide