SPATIAL AND TRANSPATIAL NETWORKS Paola Monachesi Public spaces

SPATIAL AND TRANSPATIAL NETWORKS Paola Monachesi

Public spaces

Fusion of physical and online spaces

Physical vs. Online space as a version of the “real” world Urban space as a version of online space People: they are the link

Cities as big data producers

Goal Understand collective and interaction behaviour of city/buildings’ inhabitants both online and offline. Focus is on people’s social ties Kostakos and Venkatanathan (2010) Making friends in life and online: Equivalence, microcorrelation and value in spatial and transpatial social networks. Proceedings of IEEE Social. Com, Minneapolis, USA, pp. 587 -594

People’s ties Face to face interactions are rich communicative experience but bound to space => spatial social networks Online tools lack the richness of physical interactions but go beyond space and time => transpatial social networks Combination is a fused network => overview of people’s social engagement

Questions How do online and face-to-face networks relate to each other? Do individuals assume similar roles in each network? Do transpatial networks offer greater value than spatial networks wrt. navigation through social ties?

Data 2602 participants Co-presence data [A was co-located with B] Subset of actual physical encounters, March 2007 Facebook friendship network [A is friends with B] Recorded days after Bloetooth data collection lasted 10

System Cityware application: People’s Bluetooth-enabled devices Cityware nodes Cityware servers FB application For each registered user, the system knows Bluetooth ID and FB profile ID

Spatial and transpatial networks

Types of networks Encounter Network (Spatial network) Users linked if they were co-located during the study Facebook Network (Transpatial network) Users linked if they were friends on FB Fused Network Encounter and FB networks fused 3 types of ties: Encounter, FB and “fused”

Networks Each node represents a cohort member Links represent respective ties Blue: low betweenness Red: high betweenness

Structural characteristics Measure of structural properties Size and number of edges Density Size of the largest connected component Average number of links (degree) Longest shortest path of each network (diameter) Average shortest distance between pairs of nodes Each network’s transitivity

Values structural properties

Structural characteristics

Fused Network Blue: links resulting from physical encounters Red: links resulting from FB friendship White: links resulting from both

Links Significant effect of link type on link betweenness (p<0. 0001) In the fused network Types of links in order of importance: Encounter, FB, Fused

Average betweenness in Fused NW

Triads

Structural Characteristics

Resilience

Analysis results - Equivalence Encounter network and FB network with similar characteristics Fused: increased density and core but diameter of the core does not shorten and average path length increases Conclusion: when users adopt FB, they increase their local connectivity but globally futher away from everyone since network is larger

Analysis results - roles Similarities between encounter’s network and FB network with respect to effort to maintain the network (high correlation of degree 0. 696) Online only relationships are more likely to be weak, but unclear whether Granovetter’s work also applies to online communities

Analysis results – Value of links Links of encounter networks are more important than links of FB networks Links that exist in both networks are of least importance Spatial networks might be more important because they are better at mediating the establishment of new social ties Physical co-presence enforces trust Do you agree?

A generative model Assume a fixed number of locations and people At each location people encounter each other randomly If two people encounter each other, there is a probability that they become friends on FB People may become friends on FB even if they have not met face to face Some FB friends might visit each other People might travel to locations even if they

Model results Model is a simplified version of dynamics that generate fused networks Similarity between model and observed data Support for the methodological validity of relying on Bluetooth and Facebook proxies for spatial and transpatian network proxies.

Summary: results Bluetooth and Facebook networks exhibit very similar structural characteristics As proxies to user’s SN they reflect similar aspects Fused ties least important They are more likely with close relatives or colleagues (cf. Granovetter 1973) Spatial ties more important than transpatial ties Bluetooth has the potential to record “familiar strangers”

Fusion physical and online worlds It becomes possible to: � Keep track of how people move in physical space � Investigate the effect of movement through the digital trace they leave behind � Analyze the data which is often in natural language through language technology techniques � Formalize the information extracted � Analyze: information diffusion knowledge exchange

Another use of mobility data Can we use mobility data through smart cards (i. e. Oyster cards) in order to get insights into the cities communities? The Hidden Image of the City: Sensing Community Well-Being from Urban Mobility N. Lathia, D. Quercia, J. Crowcroft, The Computer laboratory, University of Cambridge, In Pervasive 2012. Newcastle, UK. June 18 -22, 2012.

Answer Analyze the relation between London urban flow of public transport and census based indices of the various communities (i. e. community well being) Analyze the trips made by people it can be inferred which communities they belong to Goal: monitor urban spaces

Visibility

Image of a city: London

Approach

Data Well being data: IMD Index of multiple deprivation Socially deprived communities have higher IMD Richer communities have lower IMD Oyster card data All journeys made during March 2010 Data cleaning: No bus trips, inconsistencies ~76 million journeys, by 5. 1 million users Mapping between stations and IMD scores

Geographical distribution IMD values Each circle is a station, darker values have higher IMD

Methodology

Infer familiar location Identify communities that each traveller is familiar with Entries and exits of each traveller Top 2 most visited stations (~ work-home) At least 2 trips in period of observation Inferred station must not be a major rail station

Create user visit matrix It counts the visits of each traveller to a given station Binary matrix Visit = entry-exit

Community flow matrix It represents which location community members visit Each entry counts the people who live in j and who have visited i Frequency not taken into account

Correlate IMD and flow Correlation is computed using the Pearson correlation coefficient Given a vector X and a vector Y the correlation is defined as the covariance of the two variables divided by the product of the standard deviations

Compute social equaliser index It measures the extent to which an area attracts people from areas of varying deprivation If index is high the area attracts visitors from areas of varying deprivations If index is low that people within a given area tend to flow within areas with people of similar social deprivation

Compute heterogeneity index It measures the extent to which an area attracts people from areas with similar deprivation If index is high, it attracts areas different from itself

Main results The more deprived the area the more it tends to be visited Londoners do not tend to visit communities that have deprivation scores similar to theirs Rich areas tend to attract people that come from areas of various deprivations Rich people do not tend to visit communities that are deprived Segregation effects only in deprived areas

Limitations Do not know exact home location of travelers Do not know penetration of Oyster card in various communities Do not have data about urban density Only analyze portions of the city covered by public transport

Who can use this data? Urban planners Policy makers to help make decisions Transport infrastructure