Source-linked AI summary
Coupling Human Mobility and Social Ties
Jameson L. Toole, Carlos Herrera-Yague, Christian M. Schneider, Marta C. Gonzalez
TL;DR
The paper addresses how social relationships and urban mobility interact by analyzing communication and location data from three cities. It introduces mobility similarity and predictability measures, finds stronger alignment and reconstructability among social contacts, and extends a mobility model with social movement choices.
Problem
The paper examines how social behavior and mobility patterns are related when geographic information is available in communication data.
Method
The authors analyze anonymized CDR-based social and mobility networks from three cities, introduce similarity and predictability measures, and extend a mobility model with social choices.
Results
Visitation patterns are more similar to and predictable by social contacts than strangers, correlate positively with tie strength, and reveal three groups of social ties through temporal variation.
Takeaways & Limitations
Mobility similarity and predictability connect social-network structure with spatial behavior, while GeoSim reproduces the reported empirical relationships.
Abstract
from arXiv · showhide
Studies using massive, passively data collected from communication technologies have revealed many ubiquitous aspects of social networks, helping us understand and model social media, information diffusion, and organizational dynamics. More recently, these data have come tagged with geographic information, enabling studies of human mobility patterns and the science of cities. We combine these two pursuits and uncover reproducible mobility patterns amongst social contacts. First, we introduce measures of mobility similarity and predictability and measure them for populations of users in three large urban areas. We find individuals' visitations patterns are far more similar to and predictable by social contacts than strangers and that these measures are positively correlated with tie strength. Unsupervised clustering of hourly variations in mobility similarity identifies three categories of social ties and suggests geography is an important feature to contextualize social relationships. We find that the composition of a user's ego network in terms of the type of contacts they keep is correlated with mobility behavior. Finally, we extend a popular mobility model to include movement choices based on social contacts and compare it's ability to reproduce empirical measurements with two additional models of mobility.
Data
The study uses anonymized call detail records from three cities in two industrialized countries to observe communication, locations, and urban mobility over multi-month periods.
- CDRs enable both mobility observation and construction of social networks containing millions of users.
- The dataset contains over 1 billion communication events collected over 15 months in R1 and R2 and 5 months in R3.
- Each record includes event time, anonymous caller and callee identifiers, and cell-tower information for at least the caller.
Social and Mobility Measurements
The authors represent users as social-network nodes with time-resolved location matrices, then derive visitation frequencies, unique-location counts, and mobility similarity from these data.
- Users with sufficient call volume form nodes, and regular contact between users defines social-network edges.
- Each user receives a 48 × L location matrix recording calls from each city location during typical weekday and weekend hours.Rows represent 48 hourly periods and columns represent the city’s unique cell towers.
- Summing the matrix by location and normalizing by total calls yields each user’s location-visit frequencies.
- Summing across time produces a total location vector, whose nonzero entries determine the number of unique locations visited.
- Cosine similarity compares two users’ location vectors in the city’s L-dimensional location space, weighting visit frequencies rather than set overlap alone.
B Social Network C Cosine Similarity A Mobility
The framework compares mobility patterns across social-network nodes and reconstructs one user’s location vector from others’ vectors, including time-specific similarity and predictability measures.
- C Cosine Similarity: Figure 2 derives geographic cosine similarity from users’ location vectors after tracking visits and call-based social ties.
- C Cosine Similarity: Time-specific cosine similarity records how often two users visit the same places at each hour of a typical day.
- B Social Network: A user’s location vector is projected onto the subspace spanned by vectors from a selected set F of other users.QR decomposition supplies an orthonormal basis for that subspace, producing the best approximation based on F.
- C Cosine Similarity: Predictability is the ratio between the projected vector’s magnitude and the true location vector’s magnitude.A value of 1 indicates complete reconstruction from F, whereas 0 indicates that nothing about the visits can be learned.
- A Mobility: The methods are applied to social-network and mobility data from three cities.
Correlations between social behavior and mobility
Mobility similarity and predictability are higher among social contacts than randomized users, while tie strength, shared contacts, and broader social behavior correlate with mobility patterns.
- Users with more unique contacts tend to visit more unique locations and remain more predictable because additional contacts provide more reconstruction information.
- Mobility similarity and predictability are much higher for actual social contacts than for randomized users.
- Connected nodes are on average 10 times more geographically similar than randomly selected nodes, with elevated similarity extending to three network hops.
- Stronger ties have higher average mobility similarity, although the effect subsides below contact rank 10; shared contacts also correlate positively with shared locations.The shorter observation period in R3 likely biases the tail of the tie-rank distribution.
- Users with more evenly distributed calling patterns visit more unique places and are more predictable.The reported relationships remain after controlling for call-event frequency and user degree.
Contextualizing social contacts with mobility
Temporal mobility similarity separates social contacts into persistent relationship categories and connects ego-network composition with individuals’ exploratory and predictable movement.
- Relationship categories: Three persistent mobility-similarity groups distinguish acquaintances, co-workers, and family/friends by their characteristic time-of-day patterns.Acquaintances show low similarity throughout; co-workers peak during weekday work hours; family/friends peak at nights and weekends.
- Network structure: Co-worker and family/friend subgraphs retain high clustering coefficients, whereas acquaintances have much lower clustering despite comprising nearly 70% of links.The contrast supports different relationship structures across mobility-similarity clusters.
- Implications: Mobility similarity can label functional communities in social networks as well as individual social edges.This extends relationship classification beyond individual tie descriptions to network-level organization.
- Ego-network composition: Users visiting more unique locations tend to have more acquaintance ties, while less exploratory and more predictable users tend to have more co-worker and family/friend ties.The result links mobility behavior with the composition of users’ immediate social networks.
Coupling social ties and mobility
The paper extends individual mobility modeling with social influence through the GeoSim model and evaluates it against empirical data and two alternative models. GeoSim reproduces several individual mobility patterns and uniquely reproduces observed mobility similarity and predictability, while the TF model endogenously builds social networks but misses key distributions.
- Model construction: The study extends the Song et al. mobility model with movement choices based on social contacts, assuming a static social network over the studied timescales.The resulting extension is called GeoSim and is compared with the original IM model and the TF model.
- Model calibration: ⟨α⟩= 0.2 produces a close fit to observed mobility similarity and predictability distributions.The parameter α controls the influence of social contacts; α = 0 recovers the original individual mobility model, while α = 1 makes all location choices socially influenced.
- Model construction: GeoSim lets agents choose between returning to or exploring locations, with actual location choice governed by social influence with probability α or individual preference with probability 1 −α.Socially influenced choices use contacts’ mobility similarity and location-visit frequencies.
- Model comparison: GeoSim and IM reproduce location-visit frequencies well, while TF produces a flatter distribution despite adequately reproducing exploration rates.All three models estimate more absolute locations visited, but their exploration-growth rates remain consistent with empirical data.
- Limitations and future work: The TF model uniquely builds a social network endogenously but cannot recover the empirical visit-frequency and social-mobility distributions.The paper identifies developing TF variants that reproduce both social and mobility behavior as future work.
- Model comparison: Only GeoSim reproduces empirical patterns of mobility similarity and predictability; TF’s similarity values are orders of magnitude below observed data.The TF deviation is attributed to its flatter location-visit frequency distribution.
DISCUSSION
The study links social ties and urban mobility through passive mobile data, showing that mobility similarity reveals relationship structure and correlates with users’ movement behavior. It also extends mobility modeling with social behavior and identifies implications for network, mobility, and urban applications.
- High-resolution passive data quantify how mobility similarity relates to social behavior in urban settings.The study combines mobility and social-tie information from mobile and online devices.
- Mobility is more similar among social contacts than strangers, and similarity correlates positively with tie strength.Temporal variations identify three groups of social ties, including relationship types such as coworkers or family members.
- Users with many dissimilar contacts tend to explore more locations.The composition of ego networks by social-tie type is correlated with mobility behavior.
- The findings persist across three cities in two countries.
- The GeoSim model incorporates social behavior into mobility choices and replicates empirical findings.Its results were compared with two similar mobility models.
- Social-contact information may improve missing-link prediction, information- and disease-flow modeling, and travel-demand estimates.Urban applications may capture movements previously unaccounted for by incorporating social mechanisms.
AUTHORS CONTRIBUTIONS
The authors divided responsibility across data analysis, manuscript writing, study coordination, and final publication approval.
- Jameson L. Toole led data analysis and manuscript writing, while Carlos Herrera-Yagüe and Christian M. Schneider contributed to data analysis.Marta C. González coordinated the study, and all authors approved the final publication.