A study on data analysis of spatio-temporal modeling in location-based social networks
- Publication Type:
- Thesis
- Issue Date:
- 2025
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
In today’s digital age, Location-Based Social Networks (LBSNs) have become a crucial source of data for understanding user behavior and preferences. Mining data from these platforms has emerged as a significant research focus due to its potential to study patterns in user activities and preferences. The proliferation of LBSN-based applications generates vast datasets that yield valuable practical insights in areas such as public transport analysis, route optimization, disaster management, and location recommendations. The interactive features of these platforms allow users to share interests, activities, and multimedia content, producing rich datasets. These services capture and store user information alongside real-time data, enriched with metadata, textual content, multimedia, and geo-locations, facilitating comprehensive research into various aspects of user behavior patterns.
The dissertation first introduces a three-step analytical framework that combines statistical, temporal, and spatial modeling to analyze LBSN data. Previous related analysis approaches typically focus on single application domains, such as venue popularity, this dissertation introduces a comprehensive framework incorporating the three-step modeling across multiple domains, applied through case studies with analysis of user behavior in Shanghai using from Sina Weibo (Weibo), a famous LBSN in China. To enhance the accuracy of activity pattern study, Kernel Density Estimation (KDE) is implemented for anomaly detection and employs point pattern analysis to examine venue proximity effects. This approach reveals relationships between different venue types and enables detailed investigation of specific user categories, such as patterns in individuals associated with educational institutions and restaurants, providing deeper insights into activity patterns.
To address the limitations of traditional LBSN analysis, researchers either acquire domain-specific data or manually filter through millions of records. This dissertation presents Machine Learning (ML) approaches and develops Deep Learning for Location classification (Deep-Loc) models for venue classification and prediction. The proposed ML models achieve significant accuracy of 85-93% in tourism venue prediction, while the Deep-Loc methods demonstrate 99% accuracy in venue classification. These results help in find common reasons behind the specific behavior (previously addressed in literature as “maybe” or based-on assumptions). Furthermore, the study includes a hybrid group recommendation model that integrates collaborative filtering with context-aware features. This model, validated using Gowalla data, outperforms existing methods across different metrics, establishing a robust framework for LBSN data analysis applications. The research methodology encompasses detailed documentation of the implementation stages, platform specifications, design, and is supported by comprehensive pattern analysis and empirical validation using case studies with real-world datasets.
Please use this identifier to cite or link to this item: