Source-linked AI summary
Uncovering patterns of inter-urban trip and spatial interaction from social media check-in data
Yu Liu, Zhengwei Sui, Chaogui Kang, Yong Gao
TL;DR
The paper addresses limited evidence on nationwide inter-urban mobility and its relation to spatial interaction using check-in data from China. It fits a gravity model and builds a spatial network to examine distance decay, trip displacements, and communities. Aggregate interactions follow power-law distance decay, the model reproduces exponential displacements, and detected communities are spatially connected and roughly provincial, although individual movements need not follow the aggregate pattern.
Problem
Existing studies provide limited evidence on nationwide inter-urban trips and often do not differentiate human movements across spatial scales.
Method
The study extracts nationwide inter-urban movements from check-ins, fits a gravity model, and constructs a spatially embedded network for community detection.
Results
Aggregate interactions follow a power-law distance decay, the fitted model reproduces exponential trip displacements, and network communities are spatially connected and roughly province-consistent.
Takeaways & Limitations
Spatial interaction, individual mobility, and spatially embedded networks can be examined together through empirical check-in data.
Takeaways & Limitations
Aggregate agreement does not establish that each individual’s movements follows the same distance-decay model, and more detailed trajectories are needed to separate these cases.
Abstract
from arXiv · showhide
The article revisits spatial interaction and distance decay from the perspective of human mobility patterns and spatially-embedded networks based on an empirical data set. We extract nationwide inter-urban movements in China from a check-in data set that covers half million individuals and 370 cities to analyze the underlying patterns of trips and spatial interactions. By fitting the gravity model, we find that the observed spatial interactions are governed by a power law distance decay effect. The obtained gravity model also well reproduces the exponential trip displacement distribution. However, due to the ecological fallacy issue, the movement of an individual may not obey the same distance decay effect. We also construct a spatial network where the edge weights denote the interaction strengths. The communities detected from the network are spatially connected and roughly consistent with province boundaries. We attribute this pattern to different distance decay parameters between intra-province and inter-province trips.
1. Introduction
Geo-tagged check-in records provide large-scale observations of human activity, enabling aggregate mobility patterns to be studied despite stochastic individual trajectories. Existing work spans individual mobility, social networks, and regional spatial interactions, but often does not distinguish spatial scales or examine nationwide inter-urban trips.
- Check-in records combine textual content, photos, timestamps, and locations, providing geo-tagged footprints of many individuals.The paper uses these records to study mobility at scale.
- Aggregate trajectories reveal underlying mobility patterns that may not be apparent from one person’s stochastic movement.
- Prior studies investigate individual mobility, geographical effects on social networks, and aggregate spatial interactions using check-in and related data.
- Existing studies often overlook differences between spatial scales or focus on intra-urban trips rather than nationwide inter-urban movements.
2. Background
Spatial interaction research models flows between places through distance decay, while human-mobility research studies displacement and its determinants. Spatially embedded networks extend these questions to geographically located nodes and can reveal spatially connected, administratively aligned communities.
- Spatial interaction: Spatial interactions help characterize regional structure and can be measured through passenger, migration, trade, currency, telecommunications, or co-occurrence flows.
- Distance decay effect in spatial interactions: The gravity model represents interaction as place-size effects multiplied by a distance-decay function, commonly using the power law d^-β.β captures how distance affects interaction behavior.
- Distance decay effect in spatial interactions: Place population may poorly represent repulsion or attractiveness, motivating estimation of theoretical place sizes and distance-friction functions from observed interactions.
- Human mobility patterns: Human-mobility studies use multiple trajectory sources and frequently analyze displacement distributions shaped by population, activities, geography, and distance.
- Spatially-embedded network: A spatially embedded network locates nodes geographically so distances can be measured, while edge weights represent relationships between places.
- Spatially-embedded network: Conventional community detection on spatial networks has often produced spatially connected regions that coincide with administrative units.
3. Data
The study uses a large Chinese location-based social-network dataset to extract inter-urban trajectories and evaluates them against flight flows. The data cover millions of filtered check-ins, show uneven user and city activity, and capture movement patterns that only partly overlap with air travel.
- Data description: The dataset contains approximately 521,000 users’ check-ins collected over one year from September 2011 to September 2012.Filtering fake check-ins leaves about 23,500,000 records.
- Data description: 237,000 users, or 45.6%, visited at least two cities, enabling construction of inter-urban trajectories.
- Data description: User check-in counts and visited-city counts are heavy-tailed and need not correlate because repeated check-ins may occur within one city.
- Data evaluation: a comparative approach: Check-in trips and flight trips show a weak positive relationship, with R2=0.533 across 541 city-pair flows.The comparison indicates that check-ins capture movements beyond flights while underestimating some flight trips.
- Data evaluation: a comparative approach: Among the 50 city pairs with the highest Tcij/Tfij ratios, 32 pairs, or 64%, are separated by less than 1,000km.Rail travel is a major competitor to flights within this distance range.
4. Analyses
The study links collective inter-urban interactions to gravity-model distance decay while showing that aggregate displacement patterns do not uniquely determine individual movement behavior. It also finds spatially connected network communities that roughly align with provinces, potentially reflecting different intra- and inter-province decay effects.
- Interaction network: The interaction network contains 370 cities and 15,101 edges, with edge weights following a power-law distribution.The graph density is 0.351, and its diameter is 3; the maximum observed interaction is 137,847 trips between Shanghai and Suzhou.
- Fitting the gravity model: A gravity model reaches a maximum GOF of 0.985 at β=0.8, indicating power-law distance decay governs observed inter-urban interactions.The model is fitted by searching city sizes and β values from 0.1 to 2.0 using PSO.
- Interpretation and data scope: The fitted gravity approach should use empirically estimated city masses rather than assume population directly predicts interaction strength.Check-in sampling is biased and captures only part of inter-urban movement; different interaction systems can yield different theoretical city sizes and decay parameters.
- Trip displacements: The observed inter-urban displacement distribution follows P(Δd)~exp(-αΔd), with α=0.002 and distance measured in kilometers.Synthetic trips generated from the fitted gravity model produce displacement distributions that match the observed distribution.
- Individual versus collective mobility: Aggregate gravity-model regularities do not imply that every individual follows the same distance-decay pattern.Contrasting individual trajectories can produce identical collective statistics, motivating further work with more detailed data to separate these cases.
- Network communities: Detected communities are spatially connected and roughly coincide with administrative provinces, which the authors attribute to different intra- and inter-province distance-decay effects.The proposed explanation is weaker decay for intra-province trips than for inter-province trips, but only 2,053 intra-province city pairs were available and they fit poorly.
5. Discussion and conclusions
Nationwide check-in data reveal exponential inter-urban displacements alongside gravity-model spatial interactions, while network communities are spatially connected and broadly align with provinces. Different distance-decay effects for intra- and inter-province trips help explain this regional structure.
- Discussion and conclusions: Inter-urban displacements follow an exponential distribution rather than a heavy-tailed distribution.The paper relates this difference to the size and location characteristics of cities.
- Discussion and conclusions: Spatial interactions reflected by check-in data are well fitted by the gravity model.
- Discussion and conclusions: Community detection regionalizes China into mostly spatially connected communities that coincide with provinces.Closer cities generally have stronger connections, encouraging them to cluster together.
- Discussion and conclusions: Different distance-decay effects for intra-province and inter-province trips make within-province interactions relatively stronger.This difference contributes to cities in the same province being classified in the same community.