GMove: Group-Level Mobility Modeling Using Geo-Tagged Social Media

Area of Science:

• Computational Social Science
• Data Science
• Urban Informatics

Background:

• Human mobility modeling is crucial for urban planning and traffic management.
• Geo-tagged social media (GeoSM) offers vast, multi-dimensional data for mobility analysis.
• Existing methods struggle with the sparsity and complexity of GeoSM data for quality mobility modeling.

Purpose of the Study:

• To develop a novel group-level mobility modeling method using GeoSM data.
• To address the challenges of data sparsity and complexity in GeoSM for mobility modeling.
• To improve the accuracy of human mobility models and location prediction.

Main Methods:

• Propose GMove, a group-level mobility modeling method.
• Alternate between user grouping and mobility modeling to leverage their intertwined nature.
• Employ an ensemble of Hidden Markov Models (HMMs) for group-level regularity.
• Utilize a text augmenter that computes keyword correlations for spatiotemporal distributions to reduce data sparsity.

Main Results:

• GMove effectively generates meaningful group-level mobility models from GeoSM data.
• The method demonstrates robust performance on two real-life datasets.
• Extensive experiments confirm the effectiveness of the proposed approach.

Conclusions:

• GMove provides a robust framework for extracting group-level mobility patterns from sparse GeoSM data.
• The integrated approach of user grouping and mobility modeling enhances model reliability.
• GMove significantly improves context-aware location prediction accuracy compared to baseline models.