Source-linked AI summary
Understanding mobility in a social petri dish
Michael Szell, Roberta Sinatra, Giovanni Petri, Stefan Thurner, Vito Latora
TL;DR
Modeling Pardus trajectories requires explaining long-term mean-square-displacement scaling beyond a simple tendency to revisit locations. The authors use the TOM model, which incorporates the temporal order of visited locations and preferences for locations. Same-cluster jumps are observed in about 20% of cases versus 3% expected, and player motion is subdiffusive with exponent ν ≈0.26.
Problem
Modeling Pardus trajectories requires explaining long-term mean-square-displacement scaling beyond a simple tendency to revisit locations.
Source: A long-term memory model · pp. 5–6Method
The authors use the TOM model, which incorporates the temporal order of visited locations and preferences for locations.
Source: A long-term memory model · pp. 5–6Results
Same-cluster jumps are observed in about 20% of cases versus 3% expected, and player motion is subdiffusive with exponent ν ≈0.26.
Source: Mobility reveals socio-economic clusters · p. 4, A long-term memory model · p. 4Takeaways & Limitations
The model captures long-term scaling properties of movement that are fundamentally relevant to epidemic spreading and traffic management.
Source: A long-term memory model · p. 6
Abstract
from arXiv · showhide
Despite the recent availability of large data sets on human movements, a full understanding of the rules governing motion within social systems is still missing, due to incomplete information on the socio-economic factors and to often limited spatio-temporal resolutions. Here we study an entire society of individuals, the players of an online-game, with complete information on their movements in a network-shaped universe and on their social and economic interactions. Such a "socio-economic laboratory" allows to unveil the intricate interplay of spatial constraints, social and economic factors, and patterns of mobility. We find that the motion of individuals is not only constrained by physical distances, but also strongly shaped by the presence of socio-economic areas. These regions can be recovered perfectly by community detection methods solely based on the measured human dynamics. Moreover, we uncover that long-term memory in the time-order of visited locations is the essential ingredient for modeling the trajectories.
A social arena: the online game Pardus
Pardus is a network-shaped game universe where players form social and economic ties and pursue varied goals. The study uses daily movement records alongside information about players’ socio-economic environment.
A social arena: the online game Pardus
- Pardus has 400 sectors connected as a sparse network, with an average path length of 11.89 and a diameter of 27.Players can form social relations and undertake trade and production in sectors.Source: A social arena: the online game Pardus · p. 2
- The developers grouped sectors into 20 clusters that function as political or socio-economic regions, where players often focus activity for long periods.Players may also visit other clusters to explore, migrate, or during extreme events such as wars.Source: A social arena: the online game Pardus · p. 2
- Players pursue varied goals—including social relations, political influence, exploration, and trade—accompanying heterogeneous mobility patterns.Source: A social arena: the online game Pardus · p. 2
- The study analyzes daily sector positions over 1,000 days together with socio-economic information about the players’ environment.The records cover active players and avoid relying on proxies for movement or aggregates of different transport types.Source: A social arena: the online game Pardus · p. 3
Basic features of the motion
Players’ movement has a characteristic jump distance and a broad distribution of waiting times. Individual waiting-time patterns also vary, with that heterogeneity related to players’ general activity.
Basic features of the motion
- For jumps of graph distance d ≤ 15, the distribution is well-fitted by an exponential, with a characteristic jump length λ ≈ 3.Source: Basic features of the motion · p. 3
- Waiting times between consecutive jumps follow a power law with exponent β ≈2.2.A player remains in the same sector on approximately 75% of days on average; 11 of 1,458 active players never jump during the observation period.Source: Basic features of the motion · p. 3
- Individual players’ average waiting times are power-law distributed, indicating heterogeneous motion related to heterogeneity in general activity.Source: Basic features of the motion · p. 3
Mobility reveals socio-economic clusters
Players’ jumps are shaped by socio-economic cluster borders, beyond what graph distance alone predicts. Communities detected from movement transitions recover the game’s socio-economic areas far better than communities detected from network topology.
Mobility reveals socio-economic clusters
- At graph distance d = 8, same-cluster jumps occur in about 20% of cases, versus 3% expected from the null model.The measured same-cluster jump fraction exceeds the fraction of sector pairs in the same cluster at a given distance; it declines with distance and reaches zero at d = 12.Source: Mobility reveals socio-economic clusters · pp. 3–4, Figure 3 · p. 4
- Because sectors within a cluster are near one another, the analysis tests whether topology alone explains long stays in clusters; its community-detection results show it does not.The adjacency and distance matrices have block-diagonal structure, yet communities detected from them differ significantly from the socio-economic clusters.Source: Mobility reveals socio-economic clusters · p. 4
- Community detection on the player transition-count matrix reveals almost perfectly the socio-economic areas, while methods using adjacency or distance matrices deviate significantly from them.The transition matrix carries dynamic information; the adjacency and distance matrices reflect network topology.Source: Mobility reveals socio-economic clusters · p. 4
A long-term memory model
Player mobility is subdiffusive, and topology or return frequency alone does not reproduce its long-term scaling. A model using the time order of visited locations captures that pattern, while mobility transitions also recover the socio-economic clusters.
A long-term memory model
- ν ≈0.26 characterizes player subdiffusion, which plain random walks on the same network do not reproduce.Random walks show ν ≈1 up to about 100 days, followed by rapid saturation.Source: A long-term memory model · p. 4
- The transition-count matrix M recovers socio-economic clusters far better than network adjacency or shortest-path distances.Detected communities achieve F = 0.96 for M, compared with F = 0.68 for adjacency and F = 0.49 for distance.Source: Figure 4 · p. 5
- A Markov model using day-to-day transition probabilities and a preferential-return model both fail to reproduce the empirical asymptotic MSD scaling.The Markov model uses the transition counts, while preferential return reinforces frequently visited locations.Source: A long-term memory model · pp. 4–5
- Recent visits matter more than visit frequency alone: players return more readily to locations visited recently than to heavily visited locations long ago.The empirical first-return distribution follows a power law with exponent α ≈1.3 and an exponential cutoff.Source: A long-term memory model · pp. 5–6
- The Time Order Memory model combines empirical return-time, waiting-time, and jump-distance distributions to reproduce subdiffusion with ν_TOM = 0.23 ± 0.02.Its return probability is v ≈0.83; its memory of visit order allows a previously preferred location to become less frequented.Source: A long-term memory model · p. 6
Methods
The methods use daily sector records from active Artemis players to characterize movement and return patterns. They define transition probabilities, mean square displacement, and jump-time first returns for analyzing mobility.
Methods
- The Artemis data cover 1,458 active players over 1,000 days, with sector positions logged daily at 05:35 GMT.The first 200 days were excluded to avoid a possible transient phase in the developing society.Source: Methods · p. 7
- The transition count M records player movements from sector i to sector j on consecutive days, and Π normalizes these counts into transition probabilities.For each origin sector, the probabilities across destination sectors sum to one.Source: Methods · p. 7
- Mean square displacement σ2(t) averages squared distances between a player's sectors separated by t days across players and time windows.The paper classifies σ2(t) ∼ t^ν as subdiffusive for ν < 1 and super-diffusive for ν > 1.Source: Methods · p. 7
- Jump-time removes repeated stationary days from daily sector sequences, so τ counts jumps rather than elapsed days.The first-return distribution is computed in this jump-based sequence.Source: Methods · p. 7