[Paper Review] Correlation between social proximity and mobility similarity

Chao Fan, Yiding Liu+3 more|arXiv (Cornell University)|Sep 9, 2016

Human Mobility and Location-Based Analysis31 references3 citations

TL;DR

This study investigates the correlation between social proximity and mobility similarity using a large-scale Location-Based Social Network (LBSN) dataset. It finds that while the number of common friends does not significantly enhance mobility similarity, the diversity of common friends strongly predicts higher mobility similarity, challenging prior assumptions and offering a refined model for predicting human behavior from social network structure.

ABSTRACT

Human behaviors exhibit ubiquitous correlations in many aspects, such as individual and collective levels, temporal and spatial dimensions, content, social and geographical layers. With rich Internet data of online behaviors becoming available, it attracts academic interests to explore human mobility similarity from the perspective of social network proximity. Existent analysis shows a strong correlation between online social proximity and offline mobility similari- ty, namely, mobile records between friends are significantly more similar than between strangers, and those between friends with common neighbors are even more similar. We argue the importance of the number and diversity of com- mon friends, with a counter intuitive finding that the number of common friends has no positive impact on mobility similarity while the diversity plays a key role, disagreeing with previous studies. Our analysis provides a novel view for better understanding the coupling between human online and offline behaviors, and will help model and predict human behaviors based on social proximity.

Motivation & Objective

• To investigate the relationship between social proximity and mobility similarity in human behavior.
• To determine whether the number or diversity of common friends better predicts mobility similarity.
• To challenge the assumption that more common friends lead to higher mobility similarity.
• To provide a refined understanding of how social network structure influences offline mobility patterns.

Proposed method

• The study uses an LBSN dataset with real-time check-in records and explicit social ties to measure mobility similarity via Spatial Cosine (SCos) and common neighbor count (CN).
• A sampling method without replacement is applied to avoid bias from high-degree (hub) nodes, ensuring each individual appears only once in the sample.
• 1,000 independent samples are generated to ensure statistical robustness and reduce variance in comparisons.
• Kolmogorov-Smirnov (KS) tests with Bonferroni correction are used to assess whether mobility similarity distributions differ across levels of common neighbors.
• The analysis compares mobility similarity between friends with varying numbers and diversities of common friends, controlling for network sparsity and degree heterogeneity.

Experimental results

Research questions

• RQ1 Does the number of common friends significantly increase mobility similarity between individuals?
• RQ2 How does the diversity of common friends affect mobility similarity?
• RQ3 Is there a measurable correlation between social proximity and offline mobility patterns in LBSN data?
• RQ4 Can social network structure, particularly common neighbor diversity, predict mobility similarity better than common friend count?

Key findings

• The number of common friends has no statistically significant positive impact on mobility similarity, contradicting previous assumptions.
• Higher diversity among common friends is strongly correlated with increased mobility similarity, indicating a key role in behavioral alignment.
• The KS test results show that mobility similarity distributions for different common neighbor counts are not significantly different, supporting the null hypothesis of identical distributions.
• The diversity of common friends emerges as a more informative indicator of mobility similarity than the sheer number of shared connections.
• The findings suggest that social network structure, particularly the heterogeneity of shared connections, is a better predictor of human mobility patterns than simple connectivity metrics.