Introduction
Problem statement and related work
Our study and contributions
-
We identify compliant users with reliable social interaction (SMS) data within our 12-month study period. We construct a dynamic network of SMS interactions between 576 such users, by considering weekly network snapshots.
-
We study the evolution of the dynamic network, i.e., its global as well as local network structure, with time.
-
We examine whether individuals with certain behaviors in terms of evolving social network positions (i.e., centralities) or physical activities have certain (personality, depression, anxiety, etc.) traits.
-
In terms of network evolution, we see low similarity between school week snapshots and break (holiday) week snapshots. Focusing only on the more meaningful school week snapshots, we see high node overlaps. However, only snapshots from consecutive time periods have high edge overlaps. Yet, even though many edges form and break with time, global network properties such as the snapshots’ degree distributions remain stable.
-
On the other hand, the local centralities of 74% of all nodes significantly change with time. These users do not show any trait difference compared to time-stable users. However, if out of all the users whose centralities change with time we focus on those whose physical activities also change with time, then the resulting (21% of all) users are more likely to be introverted than time-stable users. Moreover, evolving centralities are significantly correlated (i.e., co-evolve) with evolving physical activities for 27% of all nodes. Users whose centralities and physical activities both change with time and who also have a co-evolution relationship (12% of all users) are more likely to be introverted as well as anxious compared to those users who are time-stable and do not have a co-evolution relationship.
-
Hence, our study reveals interesting links between individuals’ social network structure, health-related behaviors such as physical activity, and other (e.g., personality) traits.
Methods
Data description
Overview
SMS data
Fitbit data
Individuals’ static trait data from surveys
Data processing
Determining the time period and user pool from SMS data
Determining the user pool when studying co-evolution of social networks and physical activities
Network construction
Time interval (Δt) | Link threshold (w) |
---|---|
1 week | 1,2,3,4 |
1 month | 1,2,3,4 |
3 months | 1,3,6,9 |
1 semester | 1,3,6,9 |
6 months | 1,6,12,18 |
Examining the change of global network structure over time
Studying the evolution of global properties of network snapshots
Evaluating the fit of the dynamic network to different network models, i.e., graph families
Computing similarities between network snapshots at different time points
Examining the change of local network structure (centralities) over time
Local network centralities
Finding users whose network centralities evolve over time
Statistical significance of the number of users whose network positions evolve over time
Relationships and potential redundancies of different node centralities
Examining the change of users’ physical activities over time
Detecting co-evolution relationship between users’ network centralities and physical activities
Examining trait differences between different sets of users
Results
Dynamic network construction and basic network size and connectivity statistics
In-depth analysis of global network structure
Global properties of network snapshots | Local properties of nodes | Trait differences between different user groups |
---|---|---|
∙ All snapshots have high node overlaps, but only snapshots from consecutive time periods have high edge overlaps | ∙ Different centrality measures are complementary - they identify different users | ∙ Users in NET_T do not show any trait difference compared to the control group |
∙ Global network properties are stable over time | ∙ This results in 222 NET_T users, 88 PA_T users, and 81 NET_PA users | ∙ However, NET_T users who are also in PA_T are more likely to be introverted |
∙ The best fitting network model for all snapshots is the same (GEO) | ∙ The numbers of users in NET_T, PA_T, and NET_PA are statistically significantly high | ∙ In addition, a subset of these (NET_T ∩PA_T) users who are also in NET_PA are also more likely to be anxious |
Global network properties of snapshots are stable with time
Network snapshots at different time points belong to the same graph family
Similarities between network snapshots at different time points
Identification of users whose network centralities evolve over time
Identification of users whose physical activities evolve over time
Identification of users whose network centralities and physical activities co-evolve over time
Trait differences between the different user groups
Traits of users in NET_T, PA_T, and their subsets
Traits of users in NET_T ∩PA_T, NET_PA, and their subsets
-
Users whose both centrality values and physical activities change with time (in both NET_T and PA_T) have significantly lower extraversion scores (i.e., are more likely to be introverted).
-
Users whose physical activities but not centrality values change with time (in PA_T but not in NET_T) have significantly lower openness scores (i.e., are less likely to be open).
-
Users whose both centrality values and physical activities change with time and the two also co-evolve together (in all of NET_T, PA_T, and NET_PA) have significantly lower extraversion scores and higher anxiety scores (i.e., are more likely to be introverted and anxious).
-
Users whose both centrality values and physical activities change with time but do not co-evolve together (in NET_T and PA_T, but not in NET_PA) have significantly lower extraversion scores (i.e., are more likely to be introverted).